BE/PICS Joint Seminar: Uma Balakrishnan and Kunal Poorey, Sandia National Laboratories

These talks will be held jointly, with each talk 30 minutes each (1 hour total).

Talk 1: “Optimizing Anomaly Detection for GenAI based Digital Twins of Wearable Data” (Uma Balakrishnan)

In this presentation, we introduce a methodology that utilizes both real and synthetic datasets (digital twins) to address the uncertainties associated with anomaly detection thresholds in health data from wearables. By integrating state-of-the-art wearables using generative AI, and sophisticated anomaly detection techniques, our approach offers a precise and comprehensive understanding of potential health issues, significantly reducing the false negative rate. Enhancing real datasets with generative AI-based digital twins increases population size and achieves strong concordance in uncertainty analysis with results obtained from real data alone. This robust concordance is consistent even when applied to small village populations, showcasing the scalability and reliability of our generative algorithm. Validating synthetic users (digital twins) by comparing their statistical signatures with real datasets confirms the effectiveness of our approach. Our methodology promises to revolutionize healthcare data collection and address privacy concerns by providing a more comprehensive and reliable health assessment tool for early detection of biothreats or pandemics. Moreover, we have developed a versatile anomaly detection method based on the fourth-order moments of physiological parameters, applicable to a wide range of datasets and compatible with various healthcare data sources, including wearables. Our goal is to empower individuals and healthcare systems with advanced tools for real-time anomaly detection and enhanced health assessment, paving the way for improved public health outcomes.

Talk 2: “AI-aided Computational Methods to Overcome Challenges in Biology and Engineering” (Kunal Poorey)

Early detection of an emerging biothreat and robust supply of relevant medical counter measures, coupled with the capacity for developing novel therapeutics in response to emerging diseases, is vital for enhancing national healthcare resilience. Challenges such as supply chain disruptions, propriety formulations, drug resistance, and reliance on single-source raw materials or the products itself can hinder access to proper medical countermeasures. Today, advancements in artificial intelligence and machine learning are revolutionizing detection, diagnostics, and intervention strategies across all sectors of science and technology. Here we will discuss data science applications including Generative AI can early detect an emerging threat and accelerate drug discovery.

We have developed a cutting-edge computational strategy for the early detection of emerging biothreats such as infectious outbreaks (natural or intentional). Our approach utilizes advanced anomaly detection techniques applied to a diverse array of health datasets, including wearable technology, healthcare site data, Google search terms, and Twitter feeds. These sources are analyzed at multiple scales, from individual health to broader population health dynamics, to identify and notify anomalies at multiple resolutions.

Furthermore, we will discuss how machine learning (ML) aided drug discovery significantly reduces bottlenecks by enhancing both the speed and cost-effectiveness of the process. By using extensive datasets, ML algorithms identify potential drugs, predict interactions with biological targets, and optimize chemical properties for effective manufacturing. New drug development also aids in improving treatment efficacy and reducing development costs. There are still needs in developing “explainable” machine learning (XML) methods that enhances our understanding of structure-property relationships, aiding in the optimized design of effective and safer drugs by understanding model recommendations, identifying new drug targets, and predicting side effects. Additionally, we explore the potential of generative artificial intelligence (GenAI) in leveraging this knowledge to further enhance accelerated drug development. We introduce MIRA (Machine Intelligence for Rapid Acceleration of Drug Discovery and Repurposing), an innovative model combining GenAI tools and XML for accelerated drug discovery. MIRA integrates state-of-the-art GenAI models capable of conditionally generating drug-like molecules. And utilizing publicly available databases, we’ve compiled a database of drug compounds and their properties, such as toxicity, solubility, and permeability. This holistic approach paves a promising future in more accurate and informed drug discovery.