The Rise of Hypothesis-Driven Artificial Intelligence in Oncology

Area of Science:

• Computational Biology
• Bioinformatics
• Oncology

Background:

• Cancer involves complex cellular deregulation beyond genetics, necessitating advanced computational methods and high-dimensional data analysis.
• Conventional artificial intelligence (AI) models often lack interpretability and fail to incorporate researcher-generated scientific hypotheses, limiting their utility in cancer discovery.
• Existing AI approaches struggle to uncover the intricate etiology of cancer from large omics datasets.

Purpose of the Study:

• To introduce hypothesis-driven AI as an innovative approach for cancer research.
• To demonstrate how hypothesis-driven AI differs from conventional AI in interpreting complex cancer data.
• To highlight the potential of integrating domain knowledge and scientific hypotheses into AI algorithm design for novel cancer discoveries.

Main Methods:

• Review of hypothesis-driven AI applications in oncology.
• Exemplification of AI in tumor classification, patient stratification, gene discovery, drug response prediction, and tumor spatial organization.
• Discussion on incorporating domain knowledge and scientific hypotheses into AI algorithm design.

Main Results:

• Hypothesis-driven AI enables the uncovering of cancer's complex etiology from big omics data.
• This approach facilitates novel cancer discoveries that may be missed by conventional AI methods.
• Applications span various oncological areas, showcasing the integration of scientific knowledge into AI.

Conclusions:

• Hypothesis-driven AI holds significant promise for discovering new mechanistic and functional insights into cancer etiology.
• It offers a new paradigm for improving the interpretability and discovery potential of AI in cancer research.
• Further research is needed to address challenges in knowledge incorporation, bias reduction, and interpretability in AI algorithm design.