MEGA-ODE: Learning Biologically Structured and Navigable Continuous Perturbation Dynamics from Sparse Omics
Perturbation-omics experiments usually measure only a subset of molecular feature, intervention and time space, leaving many response trajectories, perturbation effects and disease- or differentiation-associated transitions unobserved. Here we present MEGA-ODE, a graph-constrained continuous-time framework for reconstructing sparse dynamic omics landscapes, predicting unmeasured molecular states…
Perturbation-omics experiments typically observe only a fraction of molecular features, interventions, and time points, resulting in missing response trajectories, perturbation effects, and transitions associated with disease or differentiation. MEGA-ODE, a graph-constrained continuous-time framework, is introduced to reconstruct sparse dynamic omics landscapes, predict unseen molecular states, and prioritize virtual perturbations towards specific biological goals.
The framework combines molecular-network priors, graph neural ordinary differential equations, and context-adaptive mixture-of-experts routing.
When tested on L1000 transcriptomic perturbations and CPPA proteomic drug-response data, MEGA-ODE outperformed baseline methods in predicting unseen features and novel perturbations. In SARS-CoV-2 infection time-series data, MEGA-ODE maintained competitiveness in forecasting future time points. Applying MEGA-ODE to a COVID-19 patient cohort enhanced retrospective disease-stage stratification using predicted intermediate profiles, as opposed to relying solely on observed profiles.
The expert programs identified immune and inflammatory signals linked to disease severity, while graph- and expert-level attributions in MAPK drug-response and stem-cell differentiation case studies focused on MAPK edges, developmental regulators, and transcription factor-target relationships, all backed by independent promoter-proximal ChIP-seq overlap.
MEGA-ODE specifically pinpointed candidate transcription-factor perturbations that could shift 12-36 hour profiles towards 96 hour definitive-endoderm marker signatures in hESC-to-definitive-endoderm differentiation. This approach framed trajectory navigation as a concrete hypothesis-generation task. Across the various case studies, MEGA-ODE demonstrated the potential of biologically structured continuous-time modeling for prediction, interpretation, and prioritization of virtual perturbations from sparse temporal omics data.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
