DIGITAL LIFE

AI-based tool ‘measures’ cancer aggressiveness and paves the way for new therapies

With the increase in cancer cases worldwide, the disease has become increasingly complex, challenging science in the search for advances in diagnosis and treatment. In this scenario, artificial intelligence (AI) has been an ally in prediction and case detection models. A tool developed by researchers from the Ribeirão Preto School of Medicine at the University of São Paulo (FMRP-USP) and Poland may contribute to this process.

The machine learning model has proven capable of predicting the aggressiveness of some types of tumors using specific proteins, generating an index for the degree of stemness that varies from low (zero) to high (one). As this index increases, the cancer tends to be more aggressive, resistant to medication and prone to relapse.

The degree of stemness refers to how much tumor cells resemble pluripotent stem cells, those with the ability to transform into almost all types of cells in the human body. The more the disease progresses, the less the malignant cells resemble the tissue from which they originated, self-renewing and with an undifferentiated phenotype.

To develop the tool, the scientists used datasets from the Clinical Proteomic Analysis of Tumors Consortium (CPTAC) for 11 types of cancer and developed the protein expression-based stemness index (PROTsi). More than 1,300 samples from breast, ovarian, lung (squamous cell carcinoma and adenocarcinoma), kidney, uterus, brain (pediatric and adult), head and neck, colon and pancreas were analyzed.

By integrating PROTsi with proteomic data from 207 pluripotent stem cells, the group identified proteins that drive the aggressiveness of some types of these tumors. These molecules may be potential targets for new general or specific therapies. With this, the tool also contributes to the personalization of anticancer therapy, in addition to advancing the clinical development of treatments.

The findings of the study, including the validation of the results, were published on April 17 in the scientific journal Cell Genomics.

“Many of these proteins are already targets of drugs available on the market for patients with cancer and other diseases. They can be tested in future studies based on this identification. We arrived at them by making the association between the stemness phenotype and tumor aggressiveness,” Professor Tathiane Malta, from the Multiomics and Molecular Oncology Laboratory at FMRP-USP, explained to Agência FAPESP.

Table...On the last World Cancer Day, celebrated on February 4, the World Health Organization (WHO) warned that 40 people are diagnosed with the disease every minute in the world and have to undergo oncological treatments.

A leading cause of death, tumors have been affecting the younger population the most. A study published in 2023 in BMJ Oncology pointed out that the incidence of early-onset cancer in adults under 50 years of age increased by 79% between 1990 and 2019, in addition to a 28% increase in deaths. 29 types of cancer were analyzed in 204 countries.

In Brazil, the National Cancer Institute (INCA) estimates that there will be 704 thousand new cases per year in the 2023-2025 triennium. According to the publication Estimativa 2023 – Incidência de Câncer no Brasil, the most common malignant tumors are non-melanoma skin tumors (31% of total cases), followed by female breast (10.5%), prostate (10%), colon and rectum (6.5%), lung (4.6%) and stomach (3%).

Results...In the validation process, PROTsi showed consistent performance in several data sets, clearly distinguishing stem cells from differentiated cells, with different tumors positioned at intermediate levels. In the clinical outcome, PROTsi was predictive in cases of uterine and head and neck cancer, for example.

In addition, the tool was able to better differentiate higher-grade tumors in samples of adenocarcinoma, uterus, pancreas and pediatric brain cancer. “We sought to build a model that can be applied to any cancer, but we saw that it works better for some than for others. We are leaving a data source available for future work”, says Malta.

According to the professor, the USP group continues testing other computational models to improve predictions.

https://www.sciencedirect.com/science/article/pii/S2666979X25001077?via%3Dihub

mundophone