Biologically-Informed Networks Predict Phosphoproteomic Drug Responses
Researchers Antonopoulos, Nordenstorm, and Nilsson (published March 18, 2026 in PLoS Computational Biology) extend the LEMBAS biologically informed recurrent neural network to predict time‑resolved phosphoproteomic trajectories from mass‑spectrometry timeseries. They introduce phosphosite mapping and a monotonic time mapping, and demonstrate accurate interpolation and zero‑shot drug‑response prediction on an EGF stimulation dataset, outperforming naïve and fully connected baselines. The model also reveals canonical and non‑canonical signaling effects.
Key Points
- 1Extend LEMBAS RNN to learn time‑resolved phosphoproteomic trajectories with phosphosite and time mapping.
- 2Demonstrate zero‑shot drug‑response prediction on EGF dataset, outperforming naïve and fully connected baselines.
- 3Enable inference of canonical and non‑canonical signaling interactions to prioritize experiments and hypotheses.
Scoring Rationale
Strong novelty and peer‑reviewed validation with practical zero‑shot capability; applicability mainly focused on phosphoproteomic signaling studies.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
