iiinova — where your digital self keeps evolving

Cyril Malbranke

@cyrilmalbranke.bsky.social

172 followers 760 following 8 posts

Postdoc @ EPFL. Previously @ ENS and Institut Pasteur. Protein design, Protein Language Models.

Cyril Malbranke @cyrilmalbranke.bsky.social · 21/08/2025

[6/8] 🎯 Beyond PPIs: ProteomeLM predicts gene essentiality across diverse taxa (e.g. E. coli, yeast, minimal cells), highlighting its potential for broad downstream applications.

Gene essentiality, showing performance outperrforms ESM-C, and that the prediction are good on E.coli, S. cerevisae and minimal cells

120

Cyril Malbranke @cyrilmalbranke.bsky.social · 21/08/2025

[5/8] ⚡ This allows unsupervised and supervised PPI prediction at proteome scale in minutes, several orders of magnitude faster than coevolution-based methods such as DCA. Try it here: github.com/Bitbol-Lab/P...

Barplot showing speed improvement over classical DCA methods

Number of predictions in function of recall to show performance leap from classical DCA methods to ProteomeLM on human interactome (0.73 -> 0.826 AUROC)

Performance on the D-SCRIPT dataset on four organisms for supervised PPI

120

Cyril Malbranke @cyrilmalbranke.bsky.social · 21/08/2025

[4/8] 🎯 Key finding: Attention heads spontaneously encode protein–protein interaction networks. Some heads can reach an AUC of 0.92 in discriminating interacting vs non-interacting pairs.

Heatmap plot showing that ProteomeLM attention heads can distinguish interacting vs non interacting pairs in E.coli, S. cerevisiae, H. sapiens

120

Cyril Malbranke @cyrilmalbranke.bsky.social · 21/08/2025

[2/8] 🧬 Training objective: ProteomeLM uses a custom masked language modeling task, predicting masked ESM-C representations of proteins within the proteome.

120