Pathway reconstruction with predicted PPI networks

[agitter]

Predicted PPIs have been around a long time. However, there is a wave of new prediction methods that are attempting to capitalize on the successes of machine learning for proteins (including 3D structure prediction). For example:

• Predict all PPIs in Mycoplasma genitalium with AlphaFold3 https://doi.org/10.1038/s44320-026-00189-7
• Scale that to yeast https://www.evocellnet.com/2026/03/mapping-yeast-atructural-interactome.html
• Do similar things but with language models so its faster https://doi.org/10.64898/2026.03.01.708874

We can use SPRAS to consider how pathway reconstruction interacts with these types of predicted networks. Does running on a larger, predicted yeast network give different/better pathways than experimental PPI? Does the existence of a predicted PPI network give us enough of a foothold to run pathway reconstruction on species where little to no experimental data exist?