Bytes of Brilliance: How BioHPC Enables AI To Power Medicine At UT Southwestern

Artificial intelligence (AI) fuels discoveries at an unprecedented pace across research, training, and medicine. AI can identify disease patterns, draw meaningful inference, and recommend increasingly accurate prognoses from staggering amounts of complex biomedical data. With UT Southwestern (UTSW) joining TRAIN (Trust & Responsible AI Network) along with other top US academic medical centers & Microsoft, BioHPC at UTSW is in a unique position to integrate enhanced AI capabilities in all aspects of research and medicine. 

BioHPC (biomedical high-performance computing) is UTSW’s in-house HPC service that provides the robust infrastructure and computational expertise to enable biomedical research. It was conceived by and is housed in the academic yet tech-savvy Lyda Hill Department of Bioinformatics. The team covers all aspects of HPC for users, from technical jobs like system expansion and maintenance to carrying out collaborative work on computational aspects of research and clinical projects.  From personalized medicine for neurological disorders to enabling automated scientific discovery, BioHPC's capabilities support innovation and enable groundbreaking advancements across diverse disciplines. Some projects led by Lyda Hill Department of Bioinformatics faculty that regularly implement AI and that are enabled by BioHPC infrastructure & team are described below. 

Dr. Albert Montillo recognizes the critical need for reliable biomarkers & a bird’s-eye view of conditions like neurodegeneration, depression, and autism spectrum disorders for informed prognoses of disease trajectory and accurate prediction of best-outcome treatments. He develops machine learning models to integrate complex datasets— neuroimaging, genomics, clinical, proteomics, & metabolomics data – into high core capacity models that enhance predictive accuracy while ensuring interpretability across diverse patient demographics. This improved capacity now enables revelation of causal rather than simply correlative effects. For instance, these models can accurately predict individual patient response to various medications for depression, the trajectory of symptoms for Parkinson Disease patients' years in advance, and can identify biomarkers of autism for gene targeting.  

In oncology, regions within the same tumor are often genomically distinct, which hinders tailored treatments. Dr. Satwik Rajaram, with his clinical collaborator Dr. Payal Kapur, utilizes AI to predict genomics from tumor pathology slides. Their models can accurately predict driver mutations in renal cancer genes and the genomic angiogenesis score. When these models are given clinical trial data, the readout matches current state-of-the-art genomic tests with significantly less cost. This allows oncologists to better predict tumor behavior and response to therapies, thus improving outcomes for the patient.  

Beyond clinical applications, AI is revolutionizing medical education. Dr. Andrew Jamieson and his team have developed algorithms that streamline the evaluation of medical students using simulated patient interactions, reducing grading time by 90%. This innovation not only enhances educational efficiency but also standardizes evaluation criteria, ensuring consistency and accuracy in student assessments. Moreover, under the leadership of our Chair and co-founder of BioHPC, Dr. Gaudenz Danuser, BioHPC supports the integration of AI education into medical curricula, preparing future healthcare professionals to leverage AI for improved patient care and diagnostic accuracy. 

AI can facilitate and significantly speed up scientific discovery by increasing transparency of neural network algorithms. Dr. Milo Lin works on a method called Deep Distilling that automatically discovers the rules an algorithm uses to produce its output based on its training dataset and translates them into computer code for programmers to understand. This shows how the neural network reaches its output, and users can then tweak the algorithms to perform better on unseen data. By elucidating the decision-making processes of AI systems, deep distilling not only enhances research efficiency but also holds potential as a decision support tool in clinical settings, augmenting diagnostic accuracy and treatment planning.  Thus, BioHPC integrates cutting-edge computational tools with interdisciplinary expertise and empowers UT Southwestern researchers and clinicians to address complex healthcare challenges with unprecedented precision and efficacy.

Authors & references: 
The above article was written based on our UTSW Newsroom article.
Isha Shah (Administration Intern, BioHPC) wrote the 1st draft.
Prapti Mody (Program Manager, Bioinformatics) edited and produced the final version.
ChatGPT suggested article title, which was further refined by Isha & Prapti to present version.