Machine-learning Algorithm Predicts Responses to Myeloma Treatments

Joana Carvalho, PhD avatar

by Joana Carvalho, PhD |

Share this article:

Share article via email
genes and disease progression

Inserting data from genetically similar patients into a machine-learning algorithm may help predict which multiple myeloma patients will benefit from Velcade (bortezomib) or Revlimid (lenalidomide) treatment, a study shows.

Those findings were reported in the study, “Predicting treatment benefit in multiple myeloma through simulation of alternative treatment effects,” published in Nature Communications.

The degree of heterogeneity (diversity) associated with cancer poses a serious challenge to current clinical therapies. Indeed, genetic variability of tumors dictates how cancer cells respond to treatment, which explains the reason why many therapies fail in most cases and benefit only a small subset of patients.

In addition, many of these anti-cancer treatments are associated with a series of adverse side effects that could be avoided entirely if there were tools available to predict which patients, at the time of diagnosis, would benefit the most from them.

In this study, researchers developed a new tool called simulated treatment learning (STL) to predict which patients would benefit the most from specific treatments. The approach is based on the assumption that data about patients who received different treatments, but who had genetically identical tumors, could be used to predict how a particular patient would respond to an alternative treatment.

Researchers applied STL to two gene expression datasets from patients with multiple myeloma receiving two different treatments: Velcade and Revlimid.

In this first dataset, they pooled data from three Phase 3 clinical trials — TT2 (GSE2658), TT3 (GSE2658), and H65/GMMG-HD4 (GSE19784) — and analyzed gene expression signatures from 1,913 unique predictor genes in a total of 910 patients, 407 who received Velcade and 503 who did not.

In the second dataset, researchers used data from the CoMMpass trial (NCT01454297) and analyzed 3,723 unique predictor genes in a total of 662 patients, 447 who received Revlimid as a front-line treatment and 215 who did not.

With this approach, investigators identified a subset of patients (19.8 percent) who would live twice as long without disease progression when receiving Velcade, compared to the entire population.

Also, they found a subgroup of patients in the CoMMpass trial (31.1 percent) who would benefit three times more from Revlimid — in terms of disease progression or death — than the entire population.

Altogether, these findings show, study authors said, that STL is a valuable tool that is able to predict, based on gene expression signatures, which patients are more likely to respond positively to a specific treatment, which enables a personalized approach.

“STL offers an important step towards realistic personalization of cancer medicine administration by identifying gene expression markers that can be used to determine the most effective treatment for a cancer patient at the moment of diagnosis,” the investigators wrote.

“Since our work has focused on MM [multiple myeloma], an important next step is to investigate if STL is also successful in unraveling treatment benefit for other diseases,” they added.

SkylineDX is a diagnostics company that also participated in the study. The company’s CEO, Dharminder Chahal, said in a press release that “By further developing this predictive classifier and making it available at the moment of diagnosis, we help physicians and patients with this debilitating disease in creating much needed tailored treatment plans.”