Clinical validation of a machine learning-based biomarkers signature to predict response to therapy in metastatic colorectal cancer patients

Titolo del Progetto

Clinical validation of a machine learning-based biomarkers signature to predict response to therapy in metastatic colorectal cancer patients

Codice del progetto

PNRR-MCNT2-2023-12377169

Programma

PIANO NAZIONALE DI RIPRESA E RESILIENZA – Missione M6/componente C2 – Investimento:  2.1 Valorizzazione e potenziamento della ricerca biomedica del SSN

Call / Bando

PNRR – M6/C2_CALL 2023 Full Proposal - Malattie Croniche non Trasmissibili (MCnT2) ad alto impatto sui sistemi sanitari e socioassistenziali

Settore ERC

Life Sciences

Ruolo Unict

Coordinatore Unità Operativa 4

Durata del progetto in mesi

24

Data inizio

31 Agosto 2024

Data fine

30 Agosto 2026

Costo totale

€ 1.000.000,00

Quota Unict

€ 256.000,00

Coordinatore

Istituto per lo Studio, la Prevenzione e la Rete Oncologica (ISPRO)

Responsabile per Unict

Prof.ssa Vincenza Barresi

Dipartimenti e strutture coinvolte

Dipartimento di Scienze Biomediche e Biotecnologiche

Altri partner

• Istituto per lo Studio, la Prevenzione e la Rete Oncologica (ISPRO) – Firenze
• Istituto di Ricovero e Cura a Carattere Scientifico - Istituto Mediterraneo per i Trapianti e Terapie ad Alta Specializzazione (IRCCS-ISMETT) – Palermo
• Università degli Studi di Palermo – Palermo.

Abstract: Metastatic colorectal cancer (mCRC) has high incidence and mortality. Personalized treatments for mCRC disease are today limited to a small number of drugs towards molecular targets such as anti-VEGF, anti-EGFR for RAS wild-type tumors, Encorafenib for BRAF (V600E) mutated tumors, programmed death-ligand 1/programmed cell death-1(PDL-1/PD-1) inhibitors for mismatch repair deficiency (dMMR)/microsatellite instability-high (MSI-H) tumors, and possibly KRAS tyrosine kinase inhibitors for G12C KRAS mutated tumors. Nevertheless, innovative biomarkers still need to be developed for predicting the response to therapy. Indeed, tailored treatments for mCRC have not yet completely evolved due to the variety in response to drugs. Today, artificial intelligence is considered a promising tool to develop prognostic and predictive models of treatment response to aid in clinical decision making. Evolution of genetically and/or epigenetically diverse tumor-cell populations during tumor growth and progression is considered the main factor causing tumor heterogeneity, displaying inherent functional variability in tumor propagation potential and tolerance to pharmacological treatment with targeted cancer therapeutics. Molecular information has been recently employed as inputs to learning models aimed to identify an artificial intelligence driven signature able to guide clinicians to choose chemotherapies for mCRC on an individualized basis. Nevertheless, in our opinion, a panel of molecular biomarkers might improve more predictive performance of learning models over those based on individual biomarkers.

The present project aims to perform a synthesis of public repository datasets and conduct a retrospective epidemiology study to examine a panel of molecular biomarkers, e.g. chromosomal instability, mutational profiling and whole transcription in tumor specimens of mCRC patients. Machine learning technology will then be utilized to develop and validate a predictive model capable of predicting response to chemotherapy, alone or combined with targeted therapy in mCRC patients and to classify them in responders vs. non-responders to improve personalized treatment decision processes.

The primary objective of the study will be to evaluate the effectiveness of molecular learning algorithm in the prediction of the response to chemotherapy alone or combined with targeted therapy for mCRC patients. The secondary objective will be to assess the efficacy of the algorithm in discriminating responder mCRC patients vs. non-responders mCRC patients. The response to therapy by the aforementioned signature will be investigated in (a) 2,277 CRC patients using The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) public datasets, and (b) a retrospective study where we will collect formalin-fixed paraffin-embedded (FFPE) tumor specimens from the cohort of metastatic or recurrent CRC patients who went for surgical purpose at the Surgical Unit, ISMETT over the last 10 years. The results of our project could help

Parole chiave: artificial intelligence; chemotherapy; targeted therapy; responders; radiomics; biomarkers; algorithm; colorectal cancer metastasis.