Researchers and data scientists at IBM have developed three novel algorithms aimed at uncovering the underlying biological processes that cause tumors to form and grow.
And the computing behemoth is making all three tools freely available to clinical researchers and AI developers.
The offerings are summarized in a blog post written by life sciences researcher Matteo Manica and data scientist Joris Cadow, both of whom work at an IBM research lab in Switzerland.
One tool, called PaccMann—an acronym for Prediction of anticancer compound sensitivity with Multi-modal attention-based neural networks—uses data from disparate sources to help predict how cells in diseased tissue will respond to a given drug.
Manica and Cado report that this technique bested existing algorithms at projecting the sensitivity of cancer cell lines in more than 200,000 pairs of drug-cell lines.
Further, PaccMann supplies “explainability” along with its predictions, showing which genes and molecular sections it homed in on as it was making calculations.
“This information can be used by researchers as a guide to potentially help them improve or repurpose existing drugs, as well as to develop new ones,” the authors write.
Another tool, dubbed INtERAcT for Interaction Network infErence from vectoR representATions of words, automatically pulls information on protein interactions from published scientific studies.
A noteworthy strength of this tool is its ability to tease out protein interactions in the context of specific disease states. This enables comparisons with normal interactions in healthy tissue, which could help researchers better understand disease mechanisms, Manica and Cado explain.
The third algorithm, PIMKL (pathway-induced multiple kernel learning), taps existing knowledge on cellular processes to identify molecular pathways that are important for the classification of patient groups.
“The insights on differences between patient groups provided, thanks to the interpretability of the model, could lead to better understanding of cancer progression,” the authors write.
Together, these three algorithms “demonstrate how machine learning approaches can be exploited to advance biomedical research on complex diseases such as cancer,” write Manica and Cado. “Our work also shows that it is possible to incorporate explainabilty into the algorithms, thereby reinforcing trust while also guiding the search for the underlying disease mechanisms.”
The authors underscore that their motivation in making the tools publicly available is to “maximize their positive impact in the scientific community.”
To learn about each of the three algorithms in greater detail, and for links to the open-source materials, read the full blog post.