Alter Lab

Inventors of the "eigengene," we formulate comparative spectral decompositions, physics-inspired multi-tensor generalizations of the SVD, to (i) compare and integrate any data types, of any number and dimensions, and (ii) scale with data sizes. Our models (iii) are interpretable in terms of known biology and batch effects and (iv) correctly predict novel mechanisms. Our prospective and retrospective validation of a genome-wide pattern in brain tumors proved that the models discover predictors of survival and response to treatment that are (v) the most accurate and precise, (vi) clinically actionable in the general population based upon as few as 50–100 patients, and (vii) are consistent across studies and over time. We discovered this, and patterns in lung, nerve, ovarian, and uterine tumors, in public data. Such alterations were recognized in cancer, yet all other attempts to associate them with outcome failed, establishing that our AI/ML is uniquely suited to personalized medicine.