Logan GROSENICK (Weill Cornell Medicine) gave a talk.
Short abstract:
Recent years have seen a flurry of new developments using machine learning for biological subtyping in psychiatry, promising potentially transformational improvements in individualized care by enabling data-driven personalized diagnosis, prognosis, and treatment.
Biological subtypes of major depressive disorder [1] and, more recently, autism spectrum disorder [2], have shown that distinct neurobiology may be at the root of heterogeneity in psychiatric and neurodevelopmental conditions. However, such biological subtyping methods and cross-sectional neuroimaging approaches have also raised controversy, mainly centered around their feasibility and reproducibility. In this talk, I will provide supporting evidence for robust and reproducible biological subtypes in psychiatry, offer insight as to why some groups may have failed to find them, and present a new approach [3] that improves significantly on existing methods for identifying patient subgroups in multiomics and neuroimaging data.
[1] Drysdale, Andrew T., et al. "Resting-state connectivity biomarkers define neurophysiological subtypes of depression." Nature Medicine 23.1 (2017): 28-38.
[2] Buch, Amanda M., et al. "Molecular and network-level mechanisms explaining individual differences in autism spectrum disorder." Nature Neuroscience 26.4 (2023): 650-663.
[3] Buch, Amanda M., Conor Liston, and Logan Grosenick. " Simple and scalable algorithms for cluster aware precision medicine." arXiv preprint arXiv:2211.16553 v3 (2023).