همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

Predicting drug response using omics data and artificial intelligence approach in cancer

نویسندگان :

Abdolhamid Nazerpanahi¹ Mohammad Ali Hossein Beigy² Amirhossein Keyhanipour³ Kaveh Kavousi⁴ Siavash Kavousi⁵

1- University of Tehran 2- University of Tehran 3- University of Tehran 4- tehran 5- tarbiat modares university

کلمات کلیدی :

Precision oncology،Personalized Medicine،Neural networks،Drug response prediction

چکیده :

Cancer treatments often yield varied responses among patients, highlighting the urgent need for personalized therapeutic approaches. Leveraging recent advances in artificial intelligence (AI) and systems biology, this study introduces a novel AI-driven framework designed to predict how different cancer cell lines respond to specific drugs. Using a robust dataset comprising approximately 5,627,072 data points—including cell line, genomic, transcriptomic, proteomic, and mutation variations sourced from COSMIC, GDSC, and Ensembl databases—we focused on the comprehensive analysis of 3,008 crucial genes across 933 unique cancer cell lines. Our approach employs a fuzzy vector-based method to construct a gene embedding of dimensions (1, 3008) for each cell line, where each element represents a fuzzy value calculated based on the extensive dataset. This fuzzy value indicates the relative importance of gene mutations, providing a more nuanced representation of mutation impacts on drug sensitivity than traditional binary embeddings. To model drug data, we utilized Morgan fingerprints, enabling detailed molecular characterization. By integrating multi-omics (genomic, transcriptomic, and proteomic) data, we accurately modeled each gene’s contribution, reflecting the complexity of molecular interactions and their biological significance. These refined representations were fed into artificial neural networks (ANNs) to predict therapeutic responses to targeted treatments. The results demonstrate that this gene importance-based method, combined with deep learning techniques, achieves a Pearson correlation of 81.12%, which is slightly better than recent work in the field. This work marks a significant step forward in precision oncology, offering a robust, data-driven strategy that not only achieves high predictive accuracy but also elucidates the underlying biological factors influencing drug effectiveness. Ultimately, our framework advances the promise of personalized medicine, tailoring cancer treatments to each patient’s unique molecular profile, thereby improving therapeutic efficacy and minimizing adverse effects.