Ras2Vec

Ras2Vec: Deep Learning for Cancer Mutation Representation

A deep learning model for representation learning of cancer mutations, capturing similarities between mutations based on their proximity in protein structures.

What is Ras2Vec?

Ras2Vec is a deep learning model designed specifically for learning representations of cancer mutations in proteins. It is able to encode amino acid substitutions in proteins, such as those caused by mutations in cancer, into vector representations that capture similarities between different mutations.

The key idea behind Ras2Vec is that mutations occurring close together in the 3D structure of proteins are likely to have similar downstream effects. Therefore, the model encodes information about the 3D structure of proteins and the locations of mutations within this structure. This allows it to place functionally similar mutations closer together in the learned vector space.

Ras2Vec works by propagating information between nearby residues in the protein structure during training. This allows the vector representation of each mutation to be influenced by its structural neighbors. Once trained, the model can then encode new and unseen mutations into the learned vector space and retrieve similar mutations for analysis.

A key application of Ras2Vec is in predicting driver mutations in cancer genomes. By comparing the representations of cancer mutations against known drivers, the model can determine which mutations are mostly likely to confer growth advantages to tumor cells. This provides an inexpensive and fast way to nominate candidate drivers for further experimental validation.

Ras2Vec Features

Official Links