Deeper Dive
My project was inspired by the deeply personal experience of my grandpa’s battle with prostate cancer. His stories about his lifelong work on perturbation theory in nuclear engineering, combined with my passion for computer science and AI, developed as a robotics captain and software lead for a FIRST Tech Challenge team, motivated me to explore preventative treatments for early cancer detection.
Unlike existing models that rely on population-level trends, my framework, MLPA, incorporates patient-specific variables such as gene expression profiles, BMI, and age to generate individualized cancer progression simulations. In validation against bioluminescent imaging data from leukemic mice, MLPA achieved an R² value of 0.93, with predictions falling within a 95% confidence interval of observed outcomes. However, MLPA doesn’t just predict cancer. It traces its path. Through generating iterative spatial transcriptomic data, it outperforms traditional AI models by offering detailed insights, as clinical professionals can see the simulated growth of cancer at different time points instead of just receiving a single, static number.
The MLPA framework has the potential to improve quality of life for cancer patients across multiple stages of the disease. In clinical settings, it can analyze biopsy and genomic data after screening to simulate treatment outcomes, helping oncologists select therapies tailored to each patient’s profile. For pharmaceutical research, MLPA can serve as a platform for virtual trials, modeling the effects of chemotherapy drugs on virtual patients before costly and time-intensive clinical studies. It can also assist with post-surgical recovery by predicting patient trajectories based on biomarker trends and imaging data, enabling clinicians to intervene early when complications are likely.