Stanford University
Showing 161-170 of 245 Results
-
Siamak Sorooshyari
Postdoctoral Scholar, Statistics
BioMy research lies at the intersection of AI/ML, statistics, biology, and engineering. I was initially trained as an electrical engineer, with a focus on signal processing and statistical algorithms. I then pursued my PhD in a neuroscience laboratory studying stress and the blood-brain barrier, where I gained experience with biological systems, experimental design, instrumentation, and data collection. My current work brings these perspectives together as I develop computational and statistical methods to better understand and predict biological processes.
A major focus of my research is understanding how aging affects the brain and how these changes are reflected across biological scales and measurement modalities. I have studied signals recorded from individual brain regions, communication between brain networks, and changes in functional connectivity across the lifespan. An important question in this work is whether quantitative properties of biological signals, such as monotonicity, exhibit consistent relationships with age. I am particularly interested in determining how different modalities capture changes associated with aging in both healthy and diseased systems, and what these measurements reveal about the underlying biological processes. This perspective can also provide insight into the reliability and interpretability of different recording modalities as measures of biological aging. I have recently begun extending these questions beyond the brain to the brain-gut-heart axis in healthy humans. By examining relationships and coordinated changes among measurements from multiple organs, I aim to develop a more integrated understanding of healthy aging and, ultimately, of how these relationships are altered in disease, specifically neurodegeneration. This work represents a broader effort to study aging as a multidimensional biological process rather than as a phenomenon confined to a single organ or measurement modality.
In parallel, I develop statistical methods for assessing the reliability and reproducibility of unsupervised learning results. In particular, I am interested in understanding how methodological choices - including the clustering algorithm, model parameters, and the number of clusters - can affect the conclusions drawn from noisy, high-dimensional datasets. This work has led to ERICA (evaluating replicability via iterative clustering assignments), a framework for evaluating whether clustering structure can be reproduced under repeated analyses without requiring a predefined ground truth. I am applying this framework to biological datasets, including cancer and neurodegenerative diseases, where clustering is frequently used to identify molecular or phenotypic subgroups. More broadly, this work seeks to develop rigorous statistical tools that can help distinguish reproducible structure from patterns that may depend strongly on methodological choices.
