When it comes to identifying patterns in mountains of data, human beings are no match for Artificial Intelligence (AI). In particular, a branch of AI called Machine Learning is often used to find regularities in data sets – be it for stock market analysis, image and speech recognition, or the classification of cells. To reliably distinguish cancer cells from healthy cells, a team led by Dr. Altuna Akalin, head of the Bioinformatics and Omics Data Science Platform at the Max Delbrück Center for Molecular Medicine in the Helmholtz Association (MDC), has now developed a machine learning program called “ikarus.” The program found a pattern in tumor cells that is common to different types of cancer, consisting of a characteristic combination of genes. According to the team’s paper in the journal Genome Biology, the algorithm also detected types of genes in the pattern that had never been clearly linked to cancer before.
Machine learning essentially means that an algorithm uses training data to learn how to answer certain questions on its own. It does so by searching for patterns in the data that help it to solve problems. After the training phase, the system can generalize from what it has learned in order to evaluate unknown data. “It was a major challenge to get suitable training data where experts had already distinguished clearly between ‘healthy’ and ‘cancerous’ cells,” relates Jan Dohmen, the first author of the paper.
In addition, single-cell sequencing data sets are often noisy. That means the information they contain about the molecular characteristics of individual cells is not very precise – perhaps because a different number of genes is detected in each cell, or because the samples are not always processed the same way. As Dohmen and his colleague Dr. Vedran Franke, co-head of the study, reports, they sifted through countless publications and contacted quite a few research groups in order to get adequate data sets. The team ultimately used data from lung and colorectal cancer cells to train the algorithm before applying it to data sets of other kinds of tumors.
In the training phase, ikarus had to find a list of characteristic genes which it then used to categorize the cells. “We tried out and refined various approaches,” Dohmen says. It was time-consuming work, as all three scientists relate. “The key was for ikarus to ultimately use two lists: one for cancer genes and one for genes from other cells,” Franke explains.
