MOFI links discrete multi-omics snapshots into continuous cell fate trajectories
Cell fate progression rarely unfolds in a single molecular layer. Transcriptomes shift first in one compartment, chromatin accessibility changes in another, and protein programs can lag or accelerate depending on lineage pressure, developmental timing, and environmental cues. The central problem in time-series single-cell biology is not the lack of data; it is the disconnect between paired omics measurements captured at discrete moments, with each cell measured once and the underlying developmental path still inferred rather than observed.
MOFI, short for Multi-Omics Fate Inference, was built to address that gap. The framework reconstructs continuous cell fate transitions from paired multi-omics snapshots sampled across multiple time points, while jointly modeling omics-specific state dynamics, cross-omics correspondence, and shared population growth. In practical terms, that means the model is not only trying to align RNA with ATAC or protein signals at one time point, but also trying to preserve how those layers move together across development, differentiation, and perturbation.
A dynamics-constrained framework for cross-omics alignment
At the center of the method is a simple biological premise: cell state is not static, and omics layers do not advance in lockstep. A transcription factor-driven program can open chromatin before a corresponding protein abundance shift becomes visible, while downstream protein responses can reveal lineage commitment after transcript levels begin to stabilize. MOFI imposes dynamic constraints on these relationships so that the inferred trajectory respects both temporal order and cross-omics coordination.
The framework was designed to analyze paired RNA-ATAC and RNA-protein time-series datasets, and it extends beyond snapshot matching. It also supports cross-omics perturbation response analysis and the identification of candidate transformation programs that may link transcriptional modules to protein-level effects. That is important because perturbation biology depends on more than correlation. A useful model has to distinguish a transient molecular detour from a true directional change in fate.
One sentence captures the appeal of the method: discrete profiles become a developmental movie.
Benchmark performance across simulated and experimental datasets
Across simulated benchmarks and paired time-series datasets, MOFI reconstructed continuous fate transitions, population dynamics, and cross-omics changes that matched the observed measurements. The abstract does not provide numerical error metrics, effect sizes, or benchmark rankings, so the result must be read in qualitative terms. Even so, the scope of the validation matters: the model was tested against both synthetic settings and real paired omics data, a combination that is often needed to show whether a method works under controlled assumptions and under the messier conditions of biological systems.
In kidney organoids, MOFI captured coordinated RNA-ATAC progression during nephron lineage specification. That application is biologically intuitive. Nephron development requires sequential activation of lineage programs, and chromatin accessibility is expected to track the regulatory openings that permit these programs to proceed. A model that can retain the coupling between transcriptomic change and chromatin remodeling offers a way to interpret lineage specification as a coordinated, layered process rather than a single-marker transition.
The hematopoietic example is more mechanistically expansive. In progenitor systems, small shifts in regulatory state can redirect lineage allocation, and the timing of those shifts can be decisive. MOFI predicted perturbation-responsive lineage redistribution and nominated multi-omics transformation programs linking transcriptional modules with protein-level responses. That creates a useful bridge between gene regulation and effector biology, since protein responses often define the functional endpoint of a differentiation signal more directly than RNA alone.
Kidney, blood, and brain point to broad developmental utility
The model was also applied to neural developmental systems, where it resolved gradual brain organoid maturation and interneuron lineage allocation during human cortical development. Neural differentiation is a particularly demanding setting because maturation is slow, branching decisions are layered, and multiple cell states can coexist in overlapping windows. A framework that can capture gradual maturation without collapsing intermediate states into a single averaged trajectory is likely to be valuable for developmental biology, organoid profiling, and lineage inference studies that rely on sparse time points.
Across these examples, the common thread is not a disease endpoint but a systems-level question: how do cells move from one fate to another while different molecular layers remain partially synchronized and partially offset? MOFI is positioned as a dynamics-grounded answer to that question. By integrating state change, cross-omics correspondence, and population growth in one framework, it seeks to turn paired measurements into interpretable developmental trajectories rather than static correlation maps.
A second sentence matters here: static alignment is not enough when the biology is moving.
What this means for multi-omics trajectory inference
The most relevant contribution of MOFI is conceptual as much as computational. Multi-omics studies increasingly generate paired data across time, but interpretation still often relies on side-by-side inspection of RNA, chromatin, or protein layers. That approach can miss the temporal structure of fate progression, especially when one modality changes earlier than another or when perturbation effects redistribute cell populations rather than simply shifting average expression. MOFI attempts to reconstruct the developmental path itself, not just the endpoints.
Its output may also be useful for prioritizing candidate transformation programs. In a lineage system, a gene module that tracks a chromatin opening event and later correlates with protein output is a stronger mechanistic candidate than a marker that merely rises in parallel. By embedding cross-omics perturbation response analysis into the same framework, MOFI may help identify regulatory programs that are not obvious from one modality in isolation.
The article remains a preprint, and the abstract provides no peer-reviewed validation, no external clinical translation, and no claims about therapeutic use. The open question is not whether paired multi-omics can be informative—it clearly can—but how robustly a dynamics-constrained model generalizes across systems, perturbations, and sampling schemes once it is tested beyond the datasets used in development. The GitHub repository is publicly linked, which should make independent inspection and benchmarking possible, but the method will need formal evaluation by other groups before its performance can be judged against existing trajectory inference tools.