Searcharxiv⌕ Search

arXiv · 2611.14559

Predicting transcriptional interactions with GrnPFN

Abstract

Combinatorial therapies offer opportunities to improve treatment efficacy and overcome therapeutic resistance, but their rational design requires understanding how combined interventions interact beyond their individual effects. Combined genetic and chemical perturbations can elicit nonadditive transcriptional responses, yet the vast number of possible combinations limits experimental characterization. Predicting these interactions is particularly challenging when few or no combinations have been measured. Here we present GrnPFN, a prior-data fitted network that predicts nonadditive transcriptional responses to combinations from single-perturbation profiles. Pretrained on simulated screens generated from random gene-regulatory networks, GrnPFN incorporates measured combinations through in-context learning without retraining. We evaluated GrnPFN in single-cell screens spanning CRISPR activation, RNA knockdown and drug combinations, using separate cells to construct model inputs, training targets and test targets. Without measured combinations in the target screen, GrnPFN outperformed gradient boosting trained on other screens. With limited combination measurements, it outperformed kernel ridge regression fitted to the same measured pairs. These advantages extended to an additional screen combining CRISPR interference with a drug. Our results demonstrate that pretraining on simulated regulatory networks enables prediction of nonadditive transcriptional interactions across perturbation modalities when experimental measurements are scarce.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Haihui Zhang. 2026-11-05. Predicting transcriptional interactions with GrnPFN. https://searcharxiv.com/abs/2611.14559

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Molecular memory reduces timing noise in stochastic gene expression

Fluctuations in gene expression affect the timing precision of downstream events triggered when a gene product reaches a critical threshold. The first-passage time (FPT) to this threshold characterizes the timing of such events. In conventional single-step birth--death models, reaction waiting times are exponentially distributed and hence memoryless. In this setting, FPT noise exhibits a U-shaped dependence on the threshold, attaining its minimum near one-half of the steady-state mean. However, mRNA production and degradation often involve multiple sequential steps, leading to hypoexponential distributions of mRNA production times and mRNA lifetimes. Here, we investigate how molecular memory arising from multistep mRNA production and degradation affects FPT statistics in bursty gene expression. We calculate FPT statistics in the small-noise limit. Both multistep mechanisms reduce FPT noise, but through distinct mechanisms. Multistep burst arrival reduces mRNA copy-number fluctuations, thereby suppressing timing noise. Multistep degradation instead reshapes transient accumulation dynamics, reducing relative timing noise even when steady-state mRNA noise increases. The threshold that minimizes FPT noise remains close to one-half of the steady-state mean, except for multistep burst arrival with constant burst sizes, for which it shifts to lower values. Stochastic simulations support the analytical predictions within their regimes of validity and show that these qualitative conclusions persist beyond the small-noise approximation.

q-bio.MN↗

Feedback and Resource Competition in Incoherent Feedforward Loops: When Is Cumulative Dose Ordering Preserved?

Biochemical coupling can change the expression levels and timing predicted by an incoherent feedforward loop (IFFL) model without changing cumulative dose ordering. We derive ordering conditions using shared-production identities and Metzler comparison bounds, with tail guarantees beyond finite computation. A nonlinear example exhibits delayed slope reversal at a fixed dose, on an $O(1/η)$ timescale as pathway mismatch $η$ vanishes. At one fixed joint fit, validated bounds establish decreasing cumulative output across 20-100 ng nominal burden, from zero to full target loading, at every positive finite observation time for a translation-blocking model and an RNA-decay model. Full loading at 20 ng reduces 48-hour cumulative output by 13.35% and 10.18%, respectively, while endpoint concentration in the first model changes by only +0.15%. Preserved ordering therefore does not ensure quantitative agreement.

q-bio.MN↗

Rule-based curation and validation in batch fermentation of a genome-scale metabolic model of the mycoprotein fungus Fusarium venenatum A3/5

Sustainable protein supply is a central food-system challenge, and mycoprotein from Fusarium venenatum is an established, low-impact option. Its production relies on refined glucose, and yields on lower-cost feedstocks are normally found by fermentation. Genome-scale metabolic models (GEMs) can estimate these yields before any experiment, but to our knowledge none has been published for this fungus, and models transferred from related species inherit template errors. Here we built iJL1143, a GEM of strain A3/5, from an Aspergillus niger model and curated it with rules based on homology, thermodynamics and the literature. We compared it with the uncurated Draft model in four oxygen-enriched batch fermentations on glucose and fructose. Thermodynamic directionality rules and the direction of the mitochondrial GTP/GDP carrier brought the maximum ATP yield of the Draft model within the theoretical range. As a result, with sequential sugar uptake and no fitted parameters, the mean final-biomass discrepancy fell from 20% for the Draft model to 6% for iJL1143. Sequential uptake and a fitted lag time also reduced its biomass time-course error from 4.88 to 0.42 g/L. In oxygen-enriched batch fermentation, iJL1143 can therefore estimate obtainable biomass before fermentation, which could reduce the experiments needed to develop lower-cost mycoprotein.

q-bio.MN↗