publications

2025

1. PertEval-scFM: Benchmarking Single-Cell Foundation Models for Perturbation Effect Prediction

   Aaron Wenteler, Martina Occhetta, Nikhil Branson, and 9 more authors

   In Proceedings of Machine Learning Research, Jul 2025

   Abs

   In silico modeling of transcriptional responses to perturbations is crucial for advancing our understanding of cellular processes and disease mechanisms. We present PertEval-scFM, a standardized framework designed to evaluate models for perturbation effect prediction. We apply PertEval-scFM to benchmark zero-shot single-cell foundation model (scFM) embeddings against simpler baseline models to assess whether these contextualized representations enhance perturbation effect prediction. Our results show that scFM embeddings do not provide consistent improvements over baseline models, especially under distribution shift. Additionally, all models struggle with predicting strong or atypical perturbation effects. Overall, this study provides a systematic evaluation of zero-shot scFM embeddings for perturbation effect prediction, highlighting the challenges of this task and revealing the limitations of current-generation scFMs. Our findings underscore the need for specialized models and high-quality datasets that capture a broader range of cellular states. Source code and documentation can be found at: https://github.com/aaronwtr/PertEval.

2024

1. PNAS

   AlphaFold2-based prediction of the co-condensation propensity of proteins

   Shengyu Zhang, Christine M. Lim, Martina Occhetta, and 1 more author

   Proceedings of the National Academy of Sciences, Jul 2024

   Abs

   The phenomenon of protein condensation has been associated with a variety of cellular functions and implicated in a wide range of human diseases. By developing CoDropleT, a predictive model incorporating conformational properties of proteins derived from AlphaFold2, this study introduces a tool for the prediction of proteins that make up protein condensates. This tool could be used for accelerating the finding of proteins involved in protein phase separation and for elucidating the still elusive mechanisms governing protein co-condensation. We also expect CoDropleT to contribute to future explorations of the therapeutic opportunities offered by the modulation of pathological processes generated by the dysregulation of protein condensates. The process of protein phase separation into liquid condensates has been implicated in the formation of membraneless organelles (MLOs), which selectively concentrate biomolecules to perform essential cellular functions. Although the importance of this process in health and disease is increasingly recognized, the experimental identification of proteins forming MLOs remains a complex challenge. In this study, we addressed this problem by harnessing the power of AlphaFold2 to perform computational predictions of the conformational properties of proteins from their amino acid sequences. We thus developed the CoDropleT (co-condensation into droplet transformer) method of predicting the propensity of co-condensation of protein pairs. The method was trained by combining experimental datasets of co-condensing proteins from the CD-CODE database with curated negative datasets of non-co-condensing proteins. To illustrate the performance of the method, we applied it to estimate the propensity of proteins to co-condense into MLOs. Our results suggest that CoDropleT could facilitate functional and therapeutic studies on protein condensation by predicting the composition of protein condensates.