#35 · Systems biology

Learned models infer regulatory interactions from perturbation and expression data

The common thread is inference of regulatory interactions from perturbation and expression data, with models built to expose the quantities they estimate. CisTransCell predicts zero-shot single-cell transcriptional responses to unseen genetic perturbations by combining gene function, regulatory control, and cellular context. MRA+ extends Modular Response Analysis to global perturbations and reports bootstrap and sign-stability procedures for robust gene regulatory connectivity inference. An Ising-type biophysical model with mean-field approximations infers enhancer-specific transcription factor interactions from STARR-seq and fluorescence expression data. The multiscale Bayesian framework is a signaling-model analogue, augmenting parameter inference with ML-predicted binding affinities derived from protein sequence. Collectively, the items pair predictive or mechanistic models with estimation of interpretable coupling parameters, while the signaling item extends this approach to binding-affinity inputs.