WorksComputer Science
LIFT: Latent Flow Matching for Probabilistic Cell Fate Dynamics with Single Cell Snapshots
Abstract
Lineage tracing links cell states through clonal ancestry, but does not generally identify unique source-descendant pairings. Learning from such data therefore requires a model of conditional outcomes, rather than only the terminal population distribution. We introduce LIFT, a latent flow framework for probabilistic cell-state generation from lineage-informed couplings. Variational inference encodes alternative outcomes in a latent space, while a conditional prior enables sampling from the initial state alone. A Mean Flow decoder predicts future cell state directly, requiring one decoder evaluation per sampling, which makes it convenient to predict cell fate through large scale Monte-Carlo sampling. This construction combines a representation of one-to-many relationships with efficient sampling and an analysis of conditional prediction error. Experiments on synthetic and real datasets demonstrate LIFT's ability to recover both population distribution and cell fate probabilities.
Funding: L.Z. was supported by the National Natural Science Foundation of China (No. 12225102, T2321001, and 12288101) and the National Key Research and Development Program of China 2024YFA0919500.
Introduction
Understanding how an early cell state constrains its future fate is a central problem in developmental biology and cellular reprogramming. Single-cell RNA sequencing provides detailed molecular snapshots, but its destructive nature generally prevents repeated measurements of the same cell. Trajectory inference must therefore reconstruct developmental relationships from population snapshots and additional assumptions [13]. Lineage-tracing technologies provide an experimental source of such information. Methods such as CellTagging and LARRY recover heritable barcodes together with transcriptomic states, linking clonal ancestry to subsequent outcomes [1] [22]. These observations motivate models that use lineage information to predict not only a likely terminal distribution, but also the probabilities of cellular fates.
A shared barcode does not, however, generally determine a unique pairing between individual cells observed at different times. When several source and terminal cells belong to the same clone, their molecular correspondence remains only partially specified. We represent the true endpoint relationship by an unknown coupling and distinguish it from the lineage-informed coupling constructed for training. The prediction target is the conditional law
where is the observed initial state and is a region of terminal state space. Different source-descendant relationships can have identical marginals but imply different fate probabilities. Recovering this law requires more than matching the endpoint populations.
Probabilistic models of cellular differentiation range from fate mapping on observed states to continuous descriptions of population dynamics. Palantir estimates terminal fate probabilities from transitions on the transcriptomic manifold [15], while CellRank 2 integrates multiple sources of directional and temporal information through Markov state modeling [21]. Waddington-OT estimates probabilistic couplings between successive populations while accounting for growth and death [14], and PRESCIENT learns stochastic cell-state dynamics from expression time series [24]. These approaches already represent heterogeneous outcomes and, in several cases, provide fate probabilities. Without explicit lineage constraints, the inferred fate relationships depend on assumptions or constraints used by each method, but they need not coincide with those supported by lineage observations.
Explicit incorporation of lineage information provides additional constraints on these relationships. LineageOT uses lineage trees to inform ancestral-state estimation and transport coupling [3]; CoSpar reconstructs transition maps from sparse clonal observations using sparsity and local coherence [20]; and moslin combines expression similarity with lineage geometry through fused Gromov–Wasserstein transport [9]. LineageVAE further uses variational latent dynamics to reconstruct historical states toward unobserved common progenitors [11]. These studies establish lineage observations as valuable information for both transition inference and state reconstruction. Nevertheless, an endpoint transition map does not uniquely specify intermediate paths. Estimating a lineage-informed coupling and constructing a continuous generator that retains its conditional structure are thus separate, complementary tasks.
Generative modeling provides tools for constructing continuous conditional models. Flow matching learns time-dependent velocity fields without numerical integration during training [10, 17], and Mean Flow addresses sampling cost by learning interval-average velocities for one-step generation [4], but a deterministic flow initialized solely at the observed cell state produces only one outcome from that state. Augmented Bridge Matching addresses coupling preservation by retaining the initial sample in the conditioning [2], while Variational Rectified Flow Matching uses latent variables to represent multimodal velocity directions [5]. For lineage-informed applications, the remaining task is to combine one-to-many conditional representation with direct time-dependent sampling. This combination motivates the LIFT framework developed below.
We introduce LIFT, a latent flow framework that combines conditional representation, efficient generation, and error control for this task. Building on variational inference and conditional latent-variable modeling [7] [16], an encoder infers latent codes from training pairs, while a conditional prior predicts their distribution from the initial state alone. We formulate Flow Matching and Mean Flow reconstruction surrogates and use the Mean Flow decoder for direct generation. Its average velocity parameterization [4] allows only one decoder evaluation per sampling, and keeping the latent code fixed across time defines a continuous generated path. The coupling can incorporate geometric or clonal information without changing the sampling architecture.
Our theoretical analysis makes the conditional prediction objective explicit. We characterize the best deterministic approximation by its irreducible conditional-variance error and establish reconstruction bounds for both FM and Mean Flow decoders. Under the stated regularity and prior-scale assumptions, the total conditional Wasserstein error is bounded by the discrepancy between and , reconstruction under encoder samples, and mismatch between the aggregated posterior and the sampling prior. We further translate this bound into fate-probability estimates, including the effects of classification boundaries and finite Monte Carlo sampling. These results concern a fitted model and do not identify the true coupling or a unique intermediate biological trajectory.
The experiments follow the same conditional prediction perspective. A synthetic branching system tests recovery of both population distributions and branching probabilities. Comparisons across numerical sampling budgets assess the balance between one-step efficiency and prediction accuracy. Finally, mouse hematopoiesis and direct reprogramming demonstrate the use of lineage information for cell fate prediction, with improved probability scores relative to LIFT trained with geometric couplings.
We organize this paper as follows. § 2 formulates the reconstruction problem, § 3 develops LIFT, and § 4 establishes its error estimates. § 5 presents the experiments, followed by § 6 on conclusions and future work.
Problem Formulation and Conditional Fate Reconstruction
We consider trajectory reconstruction from barcode-resolved cell populations observed at two time points. The endpoint measures are given, but the correspondence between individual source and terminal states remains unknown. We distinguish the true lineage-compatible coupling, denoted by , from the coupling used for training. Our objective is to recover the conditional transition structure of .
Barcode-resolved observations and the target dynamics
Consider two observation times, rescaled to 0 and 1, and distinct barcodes. For each barcode , let be the observed nonnegative finite measures on describing its source and terminal cell populations. Assume that these measures have finite second moments. The total population measures are
where we assume , , so that are probability measures, and becomes a probability measure as well.
The true endpoint relationship is represented by an unknown coupling with marginals and . Since barcode identity is preserved, this coupling permits no transport between different barcodes. More precisely, it admits a decomposition
where denotes the finite nonnegative measures on with the indicated marginals. Equation (2) is a block-diagonal constraint with respect to barcode labels, as all blocks with distinct labels vanish.
Define the set of lineage compatible couplings by
The observations specify this constraint set, but do not generally identify its true member . In practice, training sampling uses a constructed coupling , chosen from when the barcode constraints are enforced.
Because the two couplings have the same marginals, we can write their conditional decompositions as follows.
Here, is the true distribution of terminal states conditioned on the observed molecular state , whereas is the corresponding conditional distribution induced by the training coupling. For a measurable partition of the terminal state space into distinct fates, the true fate probabilities are
To reconstruct continuous dynamics, we seek a probability law on whose initial-terminal joint distribution is . Writing for the evaluation map, the ideal endpoint requirement is
The endpoint coupling does not uniquely determine intermediate paths; their construction therefore requires additional dynamical assumptions.
For probability measures with finite second moments, we can define the Wasserstein distance
where contains all pairings with marginals and , to evaluate the distance between two measures. For a learned path law with initial marginal , define
Assuming that the generated terminal distribution has a finite second moment, we measure the true conditional reconstruction error by
The Wasserstein triangle inequality and Minkowski’s inequality yield
where
represent two sources of . This discrepancy compares future distributions while holding the initial state fixed. In particular, matching the terminal marginal does not ensure recovery of the true conditional relationship.
Before analyzing training objectives, we examine a representational limitation that persists even if the true coupling were available.
Limitations of deterministic mappings
A deterministic state generator produces an endpoint , so its terminal conditional law is . This includes deterministic ODE models with unique solutions when the initial condition is fully specified by the observed state . Their ability to recover the target coupling is limited by the conditional variability under .
Let
denote the true conditional mean and covariance.
Proposition 2.1 (Irreducible conditional error of deterministic maps). For every measurable ,
Consequently,
with the minimum attained by .
Moreover, an exact deterministic realization
exists if and only if is a Dirac measure for -almost every .
We provide proof of the proposition in SM1. The proposition separates deterministic reconstruction error into the true conditional variability and the error in approximating the true conditional mean. Improving a deterministic model can reduce the second component, but not the first.
Crucially, this limitation is conditional rather than marginal. A deterministic flow can produce a terminal population by sending different initial states to different regions. However, it produces only one endpoint from each fixed initial state. Its fate probabilities under the prescribed partition are therefore
which cannot equal a true probability without additional randomization.
Motivated by these observations, we introduce a latent representation,
whose terminal conditional law is
Unlike a deterministic state map, this representation allows multiple future outcomes from the same observed state. The following sections develop this representation into continuous dynamics and relate its training error to the conditional reconstruction objective.
Variational Learning of Latent Cell Dynamics
We now construct a probabilistic dynamical model from the training coupling . A latent variable represents alternative outcomes from the same observed initial state, while a time-dependent decoder generates the corresponding cell-state trajectories. We first derive a variational reconstruction objective, then introduce flow matching and Mean Flow as simulation-free training strategies. Throughout this section, training expectations are taken under , whereas the true conditional reconstruction error remains defined relative to as in (7).
Variational formulation
Let be a latent variable. The generative model consists of a conditional prior and a decoder , giving
Its terminal conditional distribution is
During training, an encoder uses both endpoints to infer latent representations compatible with the observed pair. During inference, the terminal state is unavailable, and latent variables are drawn from .
We parameterize the encoder and conditional prior by diagonal Gaussian distributions,
with strictly positive standard deviations. The means and standard deviations are outputs of neural networks. Define the training joint distribution
To derive a variational objective, introduce the auxiliary observation model
where is fixed. Its conditional likelihood is
Following variational inference [7] [16], Jensen’s inequality gives
Define the reconstruction error and prior-matching penalty by
Substituting the Gaussian observation model into (16) and averaging over yields
Dropping the parameter-independent constant gives the negative evidence lower bound objective
The Gaussian model in (15) provides a likelihood-based interpretation of the reconstruction objective. The generator considered below remains (11), without additional observation noise.
Direct evaluation of can be expensive when is defined through a differential equation. We therefore seek simulation-free objectives that control this reconstruction error without solving the learned dynamics during training.
Flow matching and Mean Flow decoders
Both training strategies use endpoint pairs drawn from and latent variables drawn from the encoder. For , define the reference interpolation
The latent variable is held fixed along each reference path. The interpolation specifies a training path between the endpoints, and is not assumed to be an observed biological trajectory.
Flow matching. Flow matching trains a time-dependent velocity field by regression against velocities of prescribed reference paths, avoiding numerical integration during training [10, 17]. In our latent formulation, the decoder is defined by
with terminal map
The conditional flow-matching objective is
Its inputs and regression targets are available directly from (20). For a fixed encoder, the unrestricted squared-loss minimizer is
Conditioning on allows the regression to distinguish different outcomes that may overlap in the observed state space.
The FM loss is evaluated along the prescribed reference paths, whereas generation follows the learned ODE. Under suitable regularity assumptions, the former controls the endpoint reconstruction error of the latter. See Theorem 4.1 in § 4.
Mean Flow. Mean Flow replaces instantaneous velocity with an interval-average velocity, allowing direct prediction of finite-time displacements [4]. Using a forward-time convention, consider a trajectory satisfying and define
Consequently,
Differentiating along the trajectory with respect to the starting time gives
where all derivatives on the right are evaluated at .
We parameterize the average velocity by and use the reference velocity to construct the training target. Define the directional derivative
and the target
The Mean Flow loss is
where denotes the stop-gradient operation. The directional derivative is computed by a Jacobian vector product, without constructing a full Jacobian. Detaching the target avoids differentiating through this derivative when updating the model.
The numerical value of (26) is the squared residual of the reference-path identity:
This observation allows its value to be related to reconstruction error independently of the stop-gradient update rule.
The learned conditional path and endpoint map are
We use the Mean Flow loss as a simulation-free surrogate for the variational reconstruction term. Its control of the one-step endpoint error is established in Theorem 4.2. Thus, the resulting training objective is motivated by the variational formulation, but is not algebraically identical to the negative evidence lower bound.
Training and sampling
We use the Mean Flow decoder for direct conditional generation. In addition to the prior-matching penalty, we regularize the encoder against a standard Gaussian:
The full training objective is
where and . The KL terms are evaluated analytically for the Gaussian parameterization, while the Mean Flow term is estimated from sampled endpoint pairs, latent variables, and times.
Latent samples are obtained by reparameterization:
Algorithm 3.1 LIFT training
Require: Training coupling , batch size , weights , .
Initialize encoder , conditional prior , and Mean Flow decoder .
while the training budget is not exhausted do
Sample independently from .
for do
Evaluate and .
Sample and set .
Sample and .
Set and .
Evaluate .
Compute at using a Jacobian–vector product.
Set .
Compute .
end for
Update using and the specified stop-gradient rule.
end while
return Trained parameters .
which permits joint updates of the encoder, prior, and decoder. Gradients through are retained in the direct decoder evaluation; the Mean Flow regression target is detached as in (26).
Endpoint pairs are sampled according to the prescribed coupling . When is lineage-compatible, this can be implemented by sampling barcode with probability and then sampling a pair from , for . Thus, barcode information enters through the training coupling without requiring barcode labels as network inputs.
At inference, only the conditional prior and Mean Flow decoder are required. Given an initial state , draw and evaluate
The same latent sample is used at all queried times to represent one realization of a path. A different latent sample produces another possible realization from the same initial state.
If is continuous in time, the map is continuous and satisfies . This construction defines a conditional path law. It does not require an independently trained instantaneous velocity field or numerical ODE integration.
After evaluating the conditional prior, sampling requires one decoder evaluation per realization and requested time point. For a terminal fate region , its model probability can be estimated by
This estimates ; increasing reduces Monte Carlo error but does not remove coupling or model approximation errors.
The resulting model is trained using the available coupling and sampled through the learned conditional prior. Section 4 relates its training losses to reconstruction under the unknown coupling and to the accuracy of the resulting fate-probability estimates.
Error Analysis of Conditional Dynamics and Fate Prediction
We analyze how the training objectives control conditional reconstruction and fate prediction. The analysis proceeds in three steps. First, the FM and MF losses control reconstruction under the training coupling . Second, reconstruction and latent distribution matching are related to the true conditional error under . Finally, this error is translated into fate-probability error and combined with the uncertainty of finite Monte Carlo sampling.
Throughout this section, the model parameters are fixed, and all training losses denote expectations under the prescribed training distribution rather than individual minibatch estimates. The couplings and have the same marginals and . We retain the notation
from Section 3. All proofs of the results in this section are provided in SM2.
Reconstruction guarantees for flow matching and Mean Flow
For either decoder, the training reconstruction error is
Recall that the training reference path is
The following results connect the regression losses evaluated along this reference path to the endpoint produced by the learned decoder.
Flow matching. For the FM decoder, let
and set . Its training loss is
Although this loss is evaluated on the reference interpolation, the reconstruction error concerns the solution of the learned ODE. A stability estimate bridges this difference.
Theorem 4.1 (Reconstruction control by flow matching). Assume that is measurable in time and globally -Lipschitz in its state argument, uniformly in and . Assume also that is integrable on for almost every latent variable under the training law, so that the generated ODE has a unique solution. If , then
where
The estimate accounts for the accumulation of local regression errors along the generated trajectory. It concerns the exact ODE solution; a numerical implementation introduces an additional integration error.
Mean Flow. For the Mean Flow decoder,
Define the residual along the reference path by
The numerical value of the Mean Flow loss is
The stop-gradient operation used in training changes the parameter update but not this residual value.
Theorem 4.2 (Reconstruction control by Mean Flow). Assume that, for -almost every and every , the map
is absolutely continuous on , with derivative
almost everywhere.
Assume further that is Lipschitz as a map into :
If , then
For a unified notation, define
Then, for either decoder ,
These estimates control endpoint reconstruction under . They do not by themselves establish fidelity to the unknown coupling , nor do they identify the true intermediate biological paths. The next result addresses the former question.
Total conditional reconstruction error
There are two additional discrepancies between training reconstruction and the desired conditional generation. First, the training coupling may differ from . Second, training uses the encoder , whereas sampling uses the conditional prior .
Recall the conditional decompositions
and write the conditional coupling discrepancy as
To describe the training latent distribution conditioned only on the initial state, define the aggregated posterior
Its discrepancy from the sampling prior is
The generated terminal kernel is
Its true conditional error is
Theorem 4.3 (Total conditional reconstruction error). Assume that
for -almost every and all . Assume that the required second moments and the quantities below are finite. Then
Suppose additionally that the encoder and prior are diagonal Gaussians as in (13), and that their prior standard deviations satisfy
for every latent coordinate and -almost every . Then
where
Consequently, under the assumptions of the corresponding reconstruction theorem,
The three terms in (45) have distinct interpretations. The coupling discrepancy measures the difference between the training relationships and the unknown target. The reconstruction term measures how accurately the decoder reproduces training endpoints from encoder samples. The final term measures the effect of replacing the training latent distribution with the prior used for generation.
These are conditional error guarantees for a fitted model. For later use, denote the upper bound in (47) by
Fate-probability estimation
Let be the fixed partition of terminal state space into fates. The true and model fate probabilities are
Estimating by generated samples involves two different errors: the discrepancy between and , and the finite-sampling error in estimating .
From conditional distributions to fate probabilities. A small displacement can change a hard fate label when the terminal state lies near a classification boundary. Define
The next bound separates this boundary effect from the endpoint reconstruction error.
Proposition 4.4 (Stability of fate probabilities). For every ,
where the second inequality holds under the assumptions of Theorem 4.3.
The estimate follows the geometry of a label change: a coupled pair of endpoints can have different fates only if the true endpoint lies near a boundary or the displacement between the endpoints is sufficiently large. The radius balances these two possibilities. In particular, if , vanishing conditional reconstruction error implies vanishing average fate-probability error. Furthermore, we combine the results above with standard results of Monte-Carlo sampling[6], we can obtain the following end-to-end fate probability error estimation.
Corollary 4.5 (End-to-end fate-probability error). Suppose
Under the assumptions of Theorem 4.3, with probability at least over the Monte Carlo samples,
Experimental Validation and Lineage-Informed Applications
We evaluate LIFT on synthetic and real-world datasets. We first use a synthetic branching benchmark to assess its ability to reconstruct population distributions and branching probabilities. We then examine whether one-step generation combines computational efficiency with accurate conditional prediction. Finally, we evaluate how LIFT incorporates lineage information to predict cell fate probabilities in two real datasets. Across all experiments, we compare LIFT with baselines including conditional flow matching (CFM), Mean Flow (MF) and score matching (SM, here we use an implementation of SF2M[19]) to analyze the ability of LIFT. Detailed experimental settings, model-component ablations, hyperparameter sensitivity analyses, and additional results are provided in Supplementary Section SM3.
Recovering conditional fate distributions in a synthetic dataset
Dataset construction. We sample 1,000 source states and assign a branch by
For each source state , five targets are generated conditional on its sampled branch . Specifically, we construct , . Thus, these five targets share the same barcode label, but may have different fate labels. For , the population probability of the positive branch is
where is the standard Gaussian distribution function. This analytic reference permits evaluation beyond the particular branch labels realized in the finite training sample.
The principal experiment uses . Branch probabilities are evaluated at uniformly spaced query states along the segment from to . We define the probability error convention,
Here we choose . Population level accuracy is measured by between generated and reference terminal samples.
LIFT recovers both distribution and branching probabilities. Table 1 shows that LIFT can achieve the lowest branching probability error across all tested noise levels while maintaining competitive population reconstruction. In contrast, accurate marginal reconstruction by CFM does not translate into accurate branching probabilities, illustrating the distinction between these two objectives. Note that the CFM and MF probability errors coincide with the minimum error attainable by hard branch assignments on the evaluation grid, because deterministic methods can only provide 0-1 predictions. Figure 1(b) further shows that LIFT successfully captures the conditional branching probabilities, whereas the deterministic baselines produce hard branch assignments.
| Method | ||||||
| Method | PE | PE | PE | |||
| CFM | 0.238 | 1.370 | 0.275 | 2.678 | 0.305 | 5.002 |
| Mean Flow | 0.470 | 1.370 | 1.011 | 2.678 | 1.101 | 5.002 |
| SM | 0.463 | 0.282 | 0.478 | 1.500 | 0.460 | 3.580 |
| LIFT | 0.266 | 0.121 | 0.268 | 0.457 | 0.280 | 0.809 |
Table 1. Results on the synthetic branching benchmark. Lower values are better. PE is the sum in (54).

Figure 1. Conditional reconstruction in the branching system. (a) Source and terminal populations. (b) Predicted branch probabilities along the query segment. (c) Generated endpoint distributions at a fixed initial state as varies.
Conditional distributions at a fixed initial state. Figure 1(c) examines the generated endpoint distributions with the initial state held fixed. At low noise, the predicted outcomes are concentrated in one branch. As increases, probability mass is distributed across both branches, producing a bimodal conditional distribution. This qualitative transition illustrates that LIFT can capture changes in branching uncertainty.
Efficient one-step conditional generation
We assess the accuracy-cost trade-off on the synthetic branching benchmark by varying the number of function evaluations (NFE) for
CFM and SM. We compare these iterative configurations with one-step MF and LIFT, evaluating population reconstruction, branching probability error, and inference time at several noise levels.
Table 2: Effect of NFE on the synthetic branching benchmark. Inference times are averaged over 100 timing measurements. Lower values are better for , PE, and inference time. The best values in each metric column are highlighted in bold.
LIFT balances generation efficiency and conditional accuracy. Table 2 shows that LIFT achieves the lowest branching probability error across all tested noise levels while sampling faster than every iterative configuration considered. Its population reconstruction remains competitive, although CFM with larger integration budgets attains a smaller in the lowest-noise setting. Increasing NFE improves the marginal reconstruction of the iterative baselines but does not consistently improve their branching probabilities; in particular, the reported CFM probability error remains high as it can give only 0-1 prediction. Although unaugmented MF is slightly faster than LIFT, it has larger errors in both metrics. These results show that LIFT combines efficient one-step sampling with accurate branching probabilities. Specifically, lower inference time makes it convenient to predict fate probabilities through large-scale Monte-Carlo sampling.
| Method | NFE | Time (ms) | PE | PE | PE | |||
| CFM | 1,000 | 401.09 | 0.238 | 1.370 | 0.275 | 2.678 | 0.305 | 5.002 |
| 100 | 40.26 | 0.238 | 1.370 | 0.282 | 2.678 | 0.305 | 5.002 | |
| 10 | 3.87 | 0.270 | 1.370 | 0.367 | 2.678 | 0.369 | 5.002 | |
| SM | 1,000 | 855.93 | 0.463 | 0.282 | 0.478 | 1.500 | 0.460 | 3.580 |
| 100 | 86.74 | 0.531 | 0.382 | 0.548 | 1.556 | 0.558 | 3.619 | |
| 10 | 9.65 | 0.628 | 0.623 | 0.656 | 1.501 | 0.719 | 3.834 | |
| MF | 1 | 0.56 | 0.470 | 1.370 | 1.011 | 2.678 | 1.101 | 5.002 |
| LIFT | 1 | 0.96 | 0.266 | 0.121 | 0.268 | 0.457 | 0.280 | 0.809 |
Table 2. Effect of NFE on the synthetic branching benchmark. Inference times are averaged over 100 timing measurements. Lower values are better for , PE, and inference time. The best values in each metric column are highlighted in bold.
Lineage-informed cell fate prediction in real datasets
We evaluate LIFT on two lineage-traced scRNA-seq datasets: mouse hematopoiesis measured by LARRY [22] and direct fibroblast reprogramming measured by CellTagging [1]. The hematopoietic task predicts monocyte (Mo) and neutrophil (Neu) outcomes from Day 2 to Day 6, while the reprogramming task predicts induced endoderm progenitor (iEP) and dead-end outcomes from Day 15 to Day 28. Both datasets use two-dimensional SPRING state representations. We compare geometric and lineage-informed couplings, denoted by Geo and Lin, respectively. The lineage-informed construction uses uniform pairing within each barcode for hematopoiesis and within-clone OT for reprogramming. Population reconstruction is measured by , while fate prediction is assessed by mean probability error (MeanPE) and Pearson correlation against lineage-derived reference probabilities.
Table 3: Results on the two lineage-traced datasets. Geo and Lin indicate geometric and lineage-informed coupling variants, respectively. Here SM relies on entropic optimal transport coupling as [19], so SM does not have different couplings. Lower and MeanPE and higher Pearson correlation are better. The best value in each column is highlighted in bold.
LIFT improves fate prediction with lineage information. Table 3 shows that LIFT-Lin can achieve the lowest MeanPE and highest correlation on both datasets. Compared with LIFT-Geo, incorporating lineage information improves fate-probability prediction even though the population Wasserstein distance does not decrease. Supplying lineage-informed couplings to deterministic CFM or MF does not produce comparable probability accuracy. These results indicate that accurate fate prediction requires pairing lineage-informed couplings with a stochastic generator capable of representing one-to-many transitions.
| Method | Hematopoiesis | Hematopoiesis MeanPE | Hematopoiesis | Reprogramming | Reprogramming MeanPE | Reprogramming |
| CFM–Geo | 0.301 | 0.336 | 0.657 | 1.013 | 0.288 | 0.729 |
| CFM–Lin | 0.539 | 0.366 | 0.728 | 3.393 | 0.268 | 0.729 |
| MF–Geo | 0.370 | 0.343 | 0.643 | 1.846 | 0.261 | 0.746 |
| MF–Lin | 0.549 | 0.349 | 0.689 | 3.946 | 0.363 | 0.701 |
| SM | 0.190 | 0.296 | 0.735 | 1.191 | 0.213 | 0.763 |
| LIFT–Geo | 0.247 | 0.223 | 0.705 | 0.936 | 0.211 | 0.727 |
| LIFT–Lin | 0.269 | 0.119 | 0.803 | 0.959 | 0.191 | 0.795 |
Table 3. Results on the two lineage-traced datasets. Geo and Lin indicate geometric and lineage-informed coupling variants, respectively. Here SM relies on entropic optimal transport coupling as [19], so SM does not have different couplings. Lower and MeanPE and higher Pearson correlation are better. The best value in each column is highlighted in bold.
LIFT captures graded fate bias in hematopoiesis. Figure 2 presents mouse hematopoiesis as a representative example. In the highlighted region, the lineage-derived reference contains mixed Mo and Neu outcomes with a preference toward Mo. Hard assignments replace this mixture with a sharp partition, whereas LIFT retains intermediate probabilities with a modest Mo bias. The resulting probability map therefore captures graded lineage-associated fate bias.

Figure 2. Representative fate prediction in mouse hematopoiesis. (a) Lineage-derived reference fate information. (b) Hard fate assignments. (c) Mo fate probabilities predicted by LIFT. The highlighted region contains mixed Mo and Neu clonal outcomes with a bias toward Mo. LIFT represents this region through intermediate fate probabilities rather than exclusively binary assignments.
This qualitative agreement complements the quantitative results across both datasets.
Discussion and Conclusion
Lineage tracing links molecular states to clonal ancestry, but does not generally reveal unique cell-to-cell pairings across destructively sampled populations. This motivates a conditional-generation problem: the same observed state may be compatible with multiple future outcomes, whereas a deterministic state map assigns it a single endpoint. Consequently, reproducing the terminal population distribution alone does not ensure accurate fate probabilities. A model must also represent the one-to-many relationships supported by the available lineage information.
LIFT addresses this mismatch through a variational latent representation of conditional variability. An encoder learns latent distributions from endpoint pairs, a conditional prior predicts them from the initial state, and a Mean Flow decoder generates alternative outcomes without numerical integration. Under the stated regularity and prior-scale assumptions, we bound the true conditional distribution error by the discrepancy in the training coupling, decoder reconstruction error, and latent training-sampling mismatch. The analysis further connects these errors to fate-probability estimation, providing a theoretical basis for the conditional prediction objective.
The experiments show that LIFT combines accurate branching-probability prediction with competitive population distribution reconstruction and efficient one-step sampling. In mouse hematopoiesis and direct reprogramming, lineage-informed training improves fate prediction relative to the geometric-coupling variant. The hematopoietic example further illustrates how graded predictions represent mixed clonal outcomes instead of imposing hard assignments. Together, these results support the use of LIFT for efficient prediction of differentiation and reprogramming probabilities from molecular states and lineage information. The current framework nevertheless has several limitations. Its generative dynamics do not explicitly model unbalanced effects such as proliferation and cell death, and its predictions remain dependent on the supplied coupling and chosen state representation. Endpoint constraints also do not uniquely identify intermediate biological paths or the sources of conditional variability. Future work will extend the framework to unbalanced dynamics, incorporate additional observation times and biological constraints. These extensions could improve both the biological fidelity of generated trajectories and the reliability of fate predictions.
Appendix A. Mathematical and Computational Background
This appendix introduces the concepts connecting barcode-resolved observations, conditional generation, and fate-probability estimation.
Probabilistic modeling of cell dynamics. A probability measure describes the relative abundance of cell states at time . Population evolution can therefore be modeled as a continuous family of distributions connecting and , rather than as a single representative trajectory. Waddington-OT infers probabilistic couplings between observed populations, accounting for proliferation and death [14]. TrajectoryNet learns continuous paths using neural ordinary differential equations and optimal-transport regularization [18], while PRESCIENT models population evolution through stochastic dynamics [24]. These approaches connect snapshot observations through assumed dynamics. A coupling additionally specifies how initial and terminal states are paired; lineage information constrains these relationships [3]. The conditional law then describes the possible outcomes of a fixed initial state, which are not determined by the endpoint population measures alone.
Variational inference and variational autoencoders. Latent-variable models generate observations through auxiliary variables whose values are not directly measured. Inferring their posterior distribution is often computationally difficult. Variational inference replaces this posterior with a tractable approximation and optimizes a lower bound on the data log likelihood. Variational autoencoders (VAEs) implement this idea using neural encoder and decoder networks [7, 8]. Their objective balances reconstruction against a Kullback–Leibler (KL) divergence penalty, which compares the encoder distribution with the prior. In conditional generation [16], the model also receives an input . LIFT uses during training and samples at inference. Different latent samples allow the same initial state to generate different outcomes. The latent variable represents conditional variability, not an automatically identified biological mechanism.
Flow matching and Mean Flow. Flow matching trains a velocity field by regression against velocities of easily sampled reference paths [10, 17]. For the linear path , the reference velocity is . Training therefore avoids solving the learned differential equation, although generation with an FM decoder still requires numerical integration. Mean Flow instead learns velocity averaged over a time interval [4]. For , its average velocity along a trajectory satisfies
Here, the average is over time along a trajectory, not over alternative fates. Learning this quantity allows LIFT to predict each requested state directly, with a latent code held fixed along a sampled path.
Wasserstein distance. Wasserstein distance compares distributions by the displacement required to move probability mass from one to the other [12]. For probability measures with finite second moments,
where contains all pairings with marginals and . On finite cell populations, these pairings are nonnegative matrices with prescribed row and column sums. Unlike comparing population means, this distance measures differences in the locations and spread of probability mass. Our conditional error applies it separately at each initial state and averages the squared distances, distinguishing correct future-state distributions from terminal population matching.
Further reading. Biological observations and trajectory inference. Biddy et al. [1] introduce the CellTagging reprogramming application; Weinreb et al. [22] connect hematopoietic state and fate using LARRY. Saelens et al. [13] compare trajectory-inference methods and their assumptions.
Lineage-informed endpoint relationships. LineageOT [3] explains how lineage trees inform transport couplings, while CoSpar [20] combines sparse clonal observations with state-space structure to infer transition maps.
Latent generation and flow-based computation. Kingma and Welling [7] and Sohn et al. [16] develop variational and conditional generative models. Kingma and Welling also provide a tutorial on VAEs and variational training [8]. Lipman et al. [10] introduce flow-matching training, and Geng et al. [4] explain the average-velocity formulation underlying Mean Flow.
Optimal transport: Peyré and Cuturi provide a computational introduction to couplings, Wasserstein distances, and their applications [12].
Finite-sample probability estimates. Hoeffding [6] provides the concentration inequality used to bound the sampling error of the estimated fate probabilities.
Declarations
Authorship and Contributorship. All authors have made substantial intellectual contributions to the study conception, execution, and design of the work. All authors have read and approved the final manuscript. In addition, the following contributions occurred: Conceptualization: B. Yang; Methodology: B. Yang, J. Ying, and Y. Wang; Formal analysis and investigation: B. Yang, J. Ying, and Y. Wang; Writing – original draft preparation: B. Yang; Writing – review and editing: P. Zhou and L. Zhang; Funding acquisition: L. Zhang; Resources: L. Zhang; Supervision: P. Zhou and L. Zhang.
Conflicts of Interest. The authors declare there are no conflicts of interest
Data & Code Availability. The datasets analyzed in this study are derived from previously published lineage-tracing studies, including the LARRY hematopoiesis dataset [22] and the CellTagging direct reprogramming dataset[1]. The original datasets are available from the sources cited in the corresponding publications.
The code used to generate the results presented in this study are publicly available in https://github.com/van-genshin-wander/LIFT.
Ethics. This study involves computational analyses of previously published and publicly available single-cell lineage-tracing datasets. No new experiments involving human participants or animals were conducted as part of this study.
Funding. L. Zhang was supported by the National Natural Science Foundation of China (Nos. 12225102, T2321001, and 12288101) and the National Key Research and Development Program of China (No. 2024YFA0919500).
Use of AI Tools. AI tools were used only to assist with language translation, editing, and polishing during the preparation of this manuscript. All authors have checked and verified the accuracy of all parts of this work and assume responsibility for all content.
SUPPLEMENTARY MATERIALS: LIFT: Latent Flow Matching for Probabilistic Cell Fate Dynamics with Single Cell Snapshots
Bowen Yang; Junda Ying; Yuxuan Wang; Peijie Zhou; Lei Zhang
Proof of Proposition2.1
Proof. There is only one coupling between and . Hence, by definition,
Integrating over the source marginal yields
Decompose
The cross term vanishes because
Therefore,
The first term equals , proving (9). The second term is nonnegative and vanishes for , which proves (10).
Finally, a deterministic realization has conditional law almost everywhere. Conversely, if is a Dirac measure almost everywhere, its point of concentration is , and
Proofs of the Error Estimates
We provide the proofs of the results in Section 4. Throughout the appendix, the model parameters are fixed, and
All statements about conditional distributions are understood to hold for -almost every initial state. We write
for the reference path and its velocity. Norms without a subscript are Euclidean norms. The finite second moments of the endpoint measures imply that .
Proof of Theorem 4.1
Proof. Fix a triple for which the generated ODE has a solution and the reference-path regression residual is square-integrable in time. These properties hold -almost surely under the theorem’s assumptions. Let
and define
The difference between the generated and reference paths satisfies
The state-space Lipschitz assumption gives the integral inequality
Gronwall’s inequality therefore yields
At , the Cauchy–Schwarz inequality implies
The first factor equals
Taking expectation under gives
Proof of Theorem 4.2
The proof first relates the Mean Flow residual to a direct prediction error at each terminal time. The time-regularity assumption then converts an integrated error bound into control at .
Lemma SM2.1 (Integrated Mean Flow residual). Under the absolute-continuity assumption of Theorem 4.2, define
Then, for every ,
In particular, if
then
and
Proof. Fix and a triple for which the assumed absolute continuity holds. Write
The chain rule along the reference path gives
for almost every . Consequently,
Since is absolutely continuous and
the fundamental theorem of calculus proves (SM2.2). Using and dividing by yields
The Cauchy–Schwarz inequality gives
Taking expectation proves (SM2.3). Multiplication by gives (SM2.4). These inequalities remain valid with an infinite right-hand side at exceptional times.
Proof of Theorem 4.2. Let
Because is uniform on and is uniform on , Tonelli’s theorem and Lemma SM2.1 imply
The value assigned to the conditional time distribution at has no effect on this integral. The time-regularity assumption gives
Set
The reverse triangle inequality yields
If , the conclusion is immediate. If , then is constant, and (SM2.5) gives , which is stronger than the claimed estimate. It remains to consider and .
If , integration over the interval where the lower bound in (SM2.6) is positive gives
Hence .
If , the lower bound is nonnegative throughout , and
The last inequality follows from
Thus in this case. Combining the cases proves
Sharpness of the small-loss exponent. The exponent is attained by a simple decoder satisfying the stated assumptions. Let and, for fixed and , take a scalar decoder independent of its state, latent, and starting-time arguments:
Its reference-path derivative with respect to vanishes, so . Therefore,
As with fixed, no uniformly stronger small-loss power can replace under these assumptions alone.
Proof of Theorem 4.3
We first record an elementary Gaussian comparison. The proof uses an explicit coupling and will therefore also apply after mixing encoder distributions over terminal states.
Lemma SM2.2 (Gaussian coupling and relative entropy). Let
be diagonal Gaussians on with strictly positive standard deviations. If for every coordinate, then
Proof. For a common standard Gaussian vector , set
These variables have marginals and , and their coupling cost is
This proves the first inequality.
For the second, set . The Gaussian relative entropy is
For every ,
Consequently, for each coordinate,
Summing over coordinates proves the result.
Proof of Theorem 4.3. We separate reconstruction under encoder samples from the change of latent distribution at inference.
Step 1: Reconstruction under the aggregated posterior. Define
and the conditional reconstruction cost
Then . For fixed , sample and then . The pair is a coupling of and , because the marginal law of is . It follows that
Step 2: Replacing the aggregated posterior by the prior. Fix and take any coupling of and . Pushing both components through gives a coupling of and . By the latent-variable Lipschitz assumption, its squared cost is at most times the original cost. Taking the infimum over latent couplings yields
The Wasserstein triangle inequality, followed by (SM2.8) and (SM2.9), gives
Taking the norm and applying Minkowski’s inequality proves
In particular, this is a triangle-inequality estimate, not an orthogonal decomposition of squared errors.
Step 3: Gaussian control of the latent discrepancy. Fix and sample and an independent . Define
The marginal law of is the aggregated posterior, and the marginal law of is the prior. Hence,
@bibliography where the second inequality follows from Lemma SM2.2. Integration with respect to gives
This argument does not assume that the aggregated posterior is Gaussian; it explicitly couples the mixture to the Gaussian prior.
Finally, substitute this estimate and
from the appropriate reconstruction theorem into (SM2.10). This yields
as claimed.
Proof of Proposition 4.4
Proof. Fix a fate and a radius . Recall
For any ,
Indeed, if and , the line segment joining the two points does not intersect . Membership in cannot change along this segment, so the two indicators agree. No smoothness of the boundary is required for this observation.
For fixed , let be any coupling of and . Using (SM2.11),
Taking the infimum over gives the pointwise estimate
This pointwise argument does not require selecting optimal couplings measurably as a function of .
Integrating against and using the true terminal marginal,
we obtain
The right-hand side is independent of , so the same estimate holds after taking the maximum over fates. Theorem 4.3 gives the further bound obtained by replacing with . Finally, is closed because it is a finite union of boundaries. Therefore as , and
If , first choose so that the boundary mass is small, and then let the conditional reconstruction error tend to zero with this fixed. This proves the stated convergence of the average fate probabilities.
Proof of Monte-Carlo error bound
Proposition SM2.3 (Monte Carlo estimation error). For each and , the estimator
is unbiased for and satisfies
Moreover, for any , with probability at least ,
where
Proof. Condition on the fitted model and the query states. For fixed and , set
The variables are independent Bernoulli variables with success probability . Thus,
and independence gives
For completeness, we give the exponential-moment argument for the required concentration bound. For a Bernoulli variable with success probability , define
Then , and
Taylor’s formula therefore implies
By independence and exponential Markov’s inequality, for and ,
Choosing gives the upper-tail bound . Applying the same argument to the negative centered sum gives
This is the Bernoulli form of Hoeffding’s inequality.
A union bound over the pairs gives
Set
The right-hand side then equals , proving the claimed simultaneous bound. Independence between indicators of different fates is not needed for this union-bound step.
Proof of Corollary 4.5
Proof. Let be the event
By Proposition SM2.3, . On this event, for every fate ,
Table SM1: Experimental datasets. Here we use low dimension representation of real datasets deduced from standard SPRING[SM23] method.
where we used and . Proposition 4.4 now implies that, on the same event and for every ,
Taking the infimum over and substituting the expression for proves
The event does not depend on . Thus, taking this infimum introduces no additional probability penalty.
Additional Experimental Details and Results
This section provides the implementation settings, numerical-integration comparisons, hyperparameter sensitivity results, and additional hematopoietic reconstruction scores. We also present qualitative comparisons of the learned fields, latent representations, and baseline fate maps. Population reconstruction and fate-probability accuracy are reported separately throughout.
Experimental setup and evaluation protocol
Data and implementation. The controlled benchmark is a two-dimensional branching system with an analytically specified branch probability. The biological applications use barcode-resolved hematopoietic data from LARRY [SM22] and direct-reprogramming data from CellTagging [SM1]. Dataset sizes and the correspondence between experimental and rescaled times are recorded in Table SM1.
| Dataset | Time interval | Source cells | Target cells |
| Synthetic branching | 0 1 | 1,000 | 5,000 |
| Hematopoiesis | Day 2Day 6 | 1144 | 6477 |
| Reprogramming | Day 15Day 28 | 1897 | 5945 |
Table SM1. Experimental datasets. Here we use low dimension representation of real datasets deduced from standard SPRING[SM23] method.
The reported LIFT implementation uses multilayer perceptrons with ReLU activations for the encoder, conditional prior, and decoder, with a two-dimensional latent variable in the principal experiments. Training uses Adam and the objective introduced in § 3.3. Our timing results were obtained on an Intel i7-13700 CPU. Dataset-specific network widths, optimization parameters, and sensitivity results are contained below.
Training couplings and comparison methods. We compare geometric endpoint couplings with lineage-informed couplings, denoted by and . The latter use shared barcodes to restrict the admissible source–target relationships; the reprogramming experiment additionally uses OT within each retained clone. During LIFT sampling, only the molecular state is provided to the conditional prior. Barcode labels and terminal states are not inputs to the sampling network.
The primary comparisons include deterministic CFM and Mean Flow, the score-matching baseline denoted SM. We also compare latent decoders sharing the encoder–prior construction: a static MLP, an FM decoder, and the Mean Flow decoder used by LIFT. These experiments distinguish changes in the coupling from changes in the generator.
Evaluation and data separation. Population reconstruction is assessed by a Wasserstein distance in the evaluation state representation. Conditional evaluation uses analytic branch probabilities in the controlled benchmark and lineage derived outcome information in the biological applications. LIFT probabilities are estimated using 1,000 independent prior samples per initial state. The terminal fate assignment rule is shared across methods and is specified by using the convex hull as fate region. Throughout evaluated datasets, convex hulls of different labeled fates do not intersect so this assignment is well-defined. As for evaluation, we randomly separate cells from 20% barcodes as test set.
LIFT implementation
The encoder, conditional prior, and Mean Flow decoder are multilayer perceptrons with ReLU activations. The encoder receives the concatenated endpoint pair and parameterizes a diagonal Gaussian latent posterior. The conditional prior parameterizes a diagonal Gaussian using alone. The decoder receives the state, latent variable, starting time, and terminal time, and returns the interval-average velocity. The networks are trained jointly with Adam. Table SM2 lists the dataset-specific architecture and optimization settings; the real-data configuration is shared by the hematopoiesis and reprogramming experiments.
| Parameter | Synthetic | Real data |
| Encoder widths | 32, 64, 32 | 32, 128, 128, 32 |
| Prior widths | 32, 64, 32 | 32, 128, 128, 32 |
| Decoder widths | 32, 64, 32 | 32, 64, 128, 64, 32 |
| Latent dimension | 2 | 2 |
| Regularization weight | 10 | 2 |
| Reconstruction weight | 1 | 10 |
| Learning rate | ||
| Batch size | 1,000 | 1,024 |
| Training iterations | 5,000 | 20,000 |
Table SM2. LIFT architecture and training parameters. Widths specify hidden layers in order. The real-data column applies to both lineage-traced scRNA-seq applications. ReLU activations and Adam optimization are used throughout.
Training pairs are sampled from the prescribed coupling , and latent variables are sampled from using reparameterization. The time variables are sampled as and . The directional derivative in the Mean Flow target is computed by a Jacobian– vector product, and the target is detached during parameter updates. The KL divergences between diagonal Gaussians are evaluated analytically. These operations follow the training algorithm in the main article.
At inference, the encoder is not used. Given an initial state , we sample and evaluate
The same latent realization is retained across requested times along one generated path. Fate probabilities are estimated from 1,000 independent latent samples per initial state. The real-data experiments use the two-dimensional SPRING representations described in the main article. Barcode information is incorporated through the coupling rather than as an input to the sampling network. Timing measurements are obtained on an Intel i7-13700 CPU.
Hyperparameter sensitivity
We vary the latent dimension , reconstruction weight , and standard-Gaussian regularization weight on the synthetic branching benchmark. One parameter is varied at a time, with the others fixed at the default values . Table SM3 gives the complete results for .
| Parameter | Value | W2 | PE | W2 | PE | W2 | PE |
| Latent dimension | 1 | 0.417 | 0.135 | 0.288 | 0.577 | 0.302 | 1.212 |
| 2 | 0.266 | 0.121 | 0.268 | 0.457 | 0.280 | 0.809 | |
| 4 | 0.277 | 0.122 | 0.306 | 0.334 | 0.379 | 1.095 | |
| Reconstruction weight | 0.5 | 0.250 | 0.179 | 0.279 | 0.537 | 0.261 | 0.936 |
| 1 | 0.266 | 0.121 | 0.268 | 0.457 | 0.280 | 0.809 | |
| 2 | 0.272 | 0.199 | 0.286 | 0.446 | 0.279 | 0.569 | |
| 4 | 0.290 | 0.207 | 0.335 | 0.431 | 0.367 | 0.670 | |
| Regularization weight | 1 | 0.367 | 0.362 | 0.294 | 0.538 | 0.309 | 1.175 |
| 3 | 0.259 | 0.198 | 0.303 | 0.367 | 0.279 | 0.840 | |
| 5 | 0.257 | 0.216 | 0.286 | 0.382 | 0.275 | 0.936 | |
| 10 | 0.266 | 0.121 | 0.268 | 0.457 | 0.280 | 0.809 |
Table SM3. Hyperparameter sensitivity on the synthetic branching benchmark. The default setting is . Lower values are better for and PE.
Table SM3: Hyperparameter sensitivity on the synthetic branching benchmark. The default setting is . Lower values are better for and PE.
These results support the default configuration as a balanced choice across the tested regimes, rather than a uniformly optimal setting.
Ablation on LIFT components
Separating representation from decoding. We next keep the encoder and conditional prior and change the decoder: a static latent MLP, a latent FM decoder, or the latent Mean Flow decoder. The FM decoder in this ablation includes and should not be confused with the unaugmented CFM baseline in Table 1.
The results in Table SM4 show a trade-off rather than uniform dominance. Latent FM gives the lowest population in all three settings, while latent Mean Flow gives the lowest
| Decoder | Time | ||||||||
| PE | PE | PE | |||||||
| Latent FM | 43.16 | 0.201 | 0.928 | 0.234 | 0.805 | 0.267 | 1.714 | ||
| Latent MLP | 0.8516 | 0.280 | 0.350 | 0.413 | 0.813 | 0.379 | 1.325 | ||
| Latent Mean Flow | 0.9607 | 0.259 | 0.198 | 0.268 | 0.457 | 0.280 | 0.809 |
Table SM4. Latent-decoder ablation. FM uses 100 Euler steps. Times are milliseconds per batch of 1,000 points on the reported CPU, averaged over 100 timing measurements. All three variants retain latent conditioning.
Table SM4: Latent-decoder ablation. FM uses 100 Euler steps. Times are milliseconds per batch of 1,000 points on the reported CPU, averaged over 100 timing measurements. All three variants retain latent conditioning.
PE and avoids iterative integration. The static MLP is slightly faster than Mean Flow but has larger probability errors. Thus, the advantage of the proposed decoder is not that one-step generation is unique to Mean Flow, but that it combines low sampling cost with accurate conditional fate prediction in these experiments.
Additional qualitative comparisons
Conditioned prediction fields. Figure SM1 displays the instantaneous velocity fields of CFM and the interval-average velocity outputs of the unaugmented Mean Flow model. Both exhibit a low-magnitude region separating the two branch-directed regimes. In Figure SM2, the location of this separating region changes with the latent code in LIFT. For a fixed code, the prediction remains deterministic, but different codes allow the same observed state to receive different predictions. This provides a qualitative illustration of latent conditioning; it is not a proof of nonintersection or of a topological property of the learned dynamics.
Figure SM1. Vector fields derived by standard conditional flow matching (a-c) and mean vector fields derived by standard mean flow (d-f).

Figure SM2. Vector fields derived by LIFT with different latent variable .
Posterior and conditional-prior representations. Figure SM3 compares posterior means and conditional-prior means for the hematopoietic application. Endpoint pairs associated with Mo and Neu outcomes occupy different regions in the posterior representation. The encoder receives the terminal state, so this separation illustrates how endpoint information is encoded, rather than an independent prediction made from the initial state. The prior representation uses the initial state alone and also shows fate-associated structure. Fate probabilities, however, are obtained by sampling the full prior and applying the decoder, not by assigning a fate to the prior mean. These visualizations characterize learned representations without identifying a latent biological mechanism.

Figure SM3. Latent representations in the hematopoietic experiment. (a) Posterior means , which use both endpoints. (b) Conditional-prior means , which use the initial state only. Colors indicate the Mo and Neu fate annotations associated with the displayed observations.
Baseline fate maps. Figure SM4 provide the complete baseline fate maps for the hematopoiesis applications. The maps compare geometric and lineage-informed couplings for CFM and Mean Flow, together with SM. They complement the numerical population and probability scores by showing where the spatial patterns of predicted fates differ. Gray regions in the baseline visualizations mark predictions outside the assigned fate regions, rather than an additional biological fate. The maps should be read together with the quantitative comparisons, since visually separated outcomes do not by themselves establish accurate conditional probabilities or irreversible fate commitment.
Figure SM4. Visualization of cell fate probabilities for other baseline methods on the hematopoiesis dataset. Colors match the main text.
References
- [1]B. A. Biddy, W. Kong, K. Kamimoto, C. Guo, S. E. Waye, T. Sun, and S. A. Morris, Single-cell mapping of lineage and identity in direct reprogramming, Nature, 564 (2018), pp. 219–224, https://doi.org/10.1038/s41586-018-0744-4.
- [2]V. De Bortoli, G.-H. Liu, T. Chen, E. A. Theodorou, and W. Nie, Augmented bridge matching, 2023, https://doi.org/10.48550/arXiv.2311.06978.
- [3]A. Forrow and G. Schiebinger, LineageOT is a unified framework for lineage tracing and trajectory inference, Nature Communications, 12 (2021), p. 4940, https://doi.org/10.1038/s41467-021-25133-1.
- [4]Z. Geng, M. Deng, X. Bai, J. Z. Kolter, and K. He, Mean flows for one-step generative modeling, in Advances in Neural Information Processing Systems, vol. 38, 2025, https://doi.org/10.52202/085713-2534.
- [5]P. Guo and A. Schwing, Variational rectified flow matching, in Proceedings of the 42nd International Conference on Machine Learning, vol. 267 of Proceedings of Machine Learning Research, PMLR, 2025, pp. 20921–20940, https://proceedings.mlr.press/v267/guo25i.html.arxiv.org/abs/2502.09616
- [6]W. Hoeffding, Probability inequalities for sums of bounded random variables, Journal of the American Statistical Association, 58 (1963), pp. 13–30, https://doi.org/10.1080/01621459.1963.10500830.
- [7]D. P. Kingma and M. Welling, Auto-encoding variational Bayes, in International Conference on Learning Representations, 2014, https://arxiv.org/abs/1312.6114.
- [8]D. P. Kingma and M. Welling, An introduction to variational autoencoders, Foundations and Trends in Machine Learning, 12 (2019), pp. 307–392, https://doi.org/10.1561/2200000056.
- [9]M. Lange, Z. Piran, M. Klein, B. Spanjaard, D. Klein, J. P. Junker, F. J. Theis, and M. Nitzan, Mapping lineage-traced cells across time points with moslin, Genome Biology, 25 (2024), p. 277, https://doi.org/10.1186/s13059-024-03422-4.

