Advancing Biomedical Research: The Integration of AI/ML in Predictive Analysis

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Jan 5 2024

In a paper published in the journal PNAS Nexus, researchers explore the dynamic integration of artificial intelligence/machine learning (AI/ML) in biomedical research, showcasing its pivotal role in predictive analysis across diverse domains.

While AI/ML offers transformative potential in understanding complex biological phenomena, the paper accentuates the imperative of addressing challenges such as inclusivity in research, the synergy between computational models and human expertise, and the standardization of clinical data. The paper heralds a transformative era poised to optimize human health through AI/ML advancements in biomedical research by delineating these challenges as opportunities for innovation.

Biomedical Research and Precision

Aligning biomedical science with precision medicine is a cornerstone in advancing societal well-being. While fundamental science is the foundation, translating laboratory discoveries into practical applications necessitates understanding multifaceted real-world influences on health.

Precision medicine, often exemplified by tailored therapeutic approaches based on individual genetic traits, has shown remarkable progress in specific patient cohorts. However, a more comprehensive approach is essential, requiring an understanding of diverse patient perspectives on treatment success and identifying factors hindering effective therapies, encompassing biological intricacies, individual clinical nuances, and socioeconomic influences.

Biomedical research, operating within a cyclical process of scientific observation and hypothesis-driven experimentation, now encounters a deluge of intricate data surpassing conventional analytical tools' capacities. Integrating AI/ML represents a paradigm shift, where interdisciplinary collaborations merge computational prowess with molecular biology, drug development, and clinical research.

Collaborative efforts yield novel algorithms adept at distilling complex datasets into predictive probabilities, amplifying the potential for groundbreaking advancements in biomedical research. Hunter and Holmes' comprehensive review delves into these evolving methods and alternatives.

ML in Clinical Practice

Integrating into clinical applications, ML revolves around deriving predictive rules for outcomes from relevant data factors. Supervised learning methods dominate medical contexts, utilizing training datasets linking desired outcomes to numerous potential factors.

Continual refinement of decision algorithms enhances accuracy as new training data enriches the model. However, the model's reliability hinges on several key aspects, notably the suitability of the chosen ML method and the training data's quality, completeness, and relevance. While neural net-based methods, like "deep learning," excel in predictions, their black-box nature raises interpretability concerns, necessitating validation through conventional randomized clinical trials for clinical adoption.

The spectrum of ML extends beyond mere associations, as causal learning methods seek to uncover direct causal relationships rather than mere correlations. These approaches, vital where randomized controlled trials pose ethical or practical constraints, delve into the causative aspects. Addressing the clinician's need for comprehensibility, "explainable AI" methods like decision trees and probabilistic graphical models offer more straightforward interpretability. More readily applicable to clinical settings, their insights and decisions bridge the gap between complex analyses and practical clinical applications.

AI/ML in Health Initiatives

Optimal ML methodologies in healthcare necessitate a comprehensive grasp of research objectives and intended outcomes. A cross-disciplinary approach is vital, leveraging expertise from various fields to access and curate datasets pivotal for algorithmic training and validation. Human expertise remains crucial in model selection and training data curation. Translating this knowledge into societal health enhancement requires us to connect lab-based methods to clinical outcomes despite significant strides in understanding genetic makeup and expression variations.

The National Institutes of Health Multi-Omics for Health and Disease Consortium is an exemplary multidisciplinary initiative that integrates AI/ML approaches to analyze intricate datasets from diverse medical conditions. This program spans six disease study sites, aiming to comprehensively profile diseases through genomics, epigenomics, transcriptomics, and more, integrating environmental, medical history, and social determinants data. The initiative prioritizes diverse ancestral backgrounds, fostering inclusive health research and tackling health disparities.

This consortium's primary objective is to forge scalable research strategies utilizing multi-omics data and AI/ML methodologies. The immense volume of intricate data will employ AI/ML analyses at various levels, from integrating diverse omics profiles to identifying therapeutic targets and predicting clinical behavior. The substantial representation from underrepresented ancestral backgrounds will enable comparisons with established datasets, offering insights crucial for addressing health disparities.

Interdisciplinary Collaboration in Healthcare

The effectiveness of AI/ML methods in healthcare research hinges on interdisciplinary collaborations, especially involving clinical care experts. These partnerships are essential; without them, research outcomes might just identify biological states that lack significance for patients and clinicians. To produce robust results, AI/ML models necessitate datasets meticulously annotated with clinical details, standardized elements defining variables, and comprehensive representation across diverse demographics.

Challenges persist in acquiring high-quality data from the clinical care landscape, mainly due to limitations in diverse representation, especially concerning age, ethnicity, and other crucial determinants of health. While clinical trial datasets offer well-characterized endpoints, their lack of diversity poses limitations. Electronic health records (EHRs) present a potential solution, yet their variability in structure and completeness hampers the creation of high-quality datasets. Initiatives like the trusted exchange framework and standard agreement (TEFCA) aim to standardize EHR data, enabling interoperability and secure information sharing across healthcare networks.

Despite advancements, extracting essential outcome variables not routinely collected in clinical settings remains challenging for AI/ML applications, demanding standardized collection methods or alternative sources like patient-reported outcomes and wearable technology data. Collaboration among AI/ML researchers, clinicians, and health equity experts becomes pivotal in recognizing and incorporating critical confounding factors for unbiased, practical analysis guiding clinical care.

Conclusion

To sum up, the more profound examination of human biology and behavior reveals increasing complexity. AI/ML approaches propel biomedical research forward, offering the means to address this intricate nature. The excitement is warranted, alongside the heightened responsibility to focus on meeting the patients' and society's needs, interpreting results accurately, and collaborating closely to unravel the complex interrelationships defining health at individual and societal levels.