Finding the most promising indications for novel treatments in oncology

Area of Science:

• Oncology
• Computational Biology
• Machine Learning

Background:

• Developing effective cancer therapies involves significant costs and complexities.
• Identifying optimal treatment indications for specific drugs, like anti-PD-1, remains a challenge.
• Real-world data holds potential for improving treatment selection.

Purpose of the Study:

• To develop a machine learning model that predicts effective cancer therapy indications.
• To leverage representation learning on real-world patient data for treatment discovery.
• To accelerate the identification of drugs like anti-PD-1 for specific cancer types.

Main Methods:

• Trained a machine learning model (INSPIRE) to rank cancer indications based on anti-PD-1 treatment relevance.
• Utilized histology-based indications and patient medical journey data for feature generation.
• Employed an embedding approach for biological relevance and reference indications.

Main Results:

• The INSPIRE model prioritized 70% of subsequent anti-PD-1 inhibitor approvals when trained on pre-clinical data.
• The approach demonstrated the ability to predict the biological relevance of treatments for specific indications.
• Successfully identified potential effective treatments by analyzing patient medical journey events.

Conclusions:

• INSPIRE can accelerate the selection and testing of effective cancer drugs.
• This approach has the potential to reduce healthcare costs and improve patient care.
• Representation learning on real-world data offers a promising avenue for precision oncology.