ORLY ALTER portrait
  • USTAR Associate Professor of Bioengineering and Human Genetics, Scientific Computing and Imaging Institute and Huntsman Cancer Institute, University of Utah
  • Chief Scientific Officer and Co-Founder, Prism AI

Education

  • Postdoctoral Fellowship, genetics, STANFORD UNIVERSITY
  • Ph.D., applied physics, STANFORD UNIVERSITY. Project: Impossibility of Determining the Quantum Wavefunction of a Single System and Fundamental Limit to External Force Detection.
  • B.Sc. magna cum laude, physics, TEL AVIV UNIVERSITY

Biography



Orly Alter is a Utah Science, Technology, and Research associate professor of bioengineering and human genetics at the Scientific Computing and Imaging Institute1  and the Huntsman Cancer Institute at the University of Utah, a scientific advisory board member of the National Cancer Institute (NCI) -DOE Cancer Moonshot collaboration on predictive oncology, and the chief scientific officer and a co-founder of Prism AI.2  Alter received her Ph.D. in applied physics at Stanford University and her B.Sc. magna cum laude in physics at Tel Aviv University. Her Ph.D. thesis, which was published by Wiley,3,4,5  is recognized as crucial to gravitational wave detection and quantum computing.6,7,8

Inventor of the "eigengene,"9,10,11,12  Alter formulates comparative spectral decompositions, physics-inspired13  multi-tensor14,15,16  generalizations17,18,19,20  of the singular value decomposition, to (i) compare and integrate any data types, of any number and dimensions, and (ii) scale with data sizes. Her models (iii) are interpretable in terms of known biology and batch effects and (iv) correctly21  predict22,23,24,25,26  previously unknown mechanisms.27,28  Her prospective and retrospective validation29,30,31,32  of a genome-wide pattern of DNA copy-number alterations in brain33,34,35,36  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. She discovered this, and patterns in lung,37,38  nerve,39  ovarian,40,41,42,43,44,45  and uterine tumors, in public data. Such alterations were recognized in cancer, yet all other attempts to associate them with outcome failed, establishing that Alter's artificial intelligence and machine learning (AI/ML) is uniquely suited to personalized medicine.

Physics of Cancer

NCI U01 CA-202144: Multi-Tensor Decompositions for Personalized Cancer Diagnostics and Prognostics1,2,3

This Physical Sciences in Oncology project is developing new mathematical frameworks to do what no others currently can, that is, create a single coherent model from multiple high-dimensional datasets, known as tensors. The frameworks – comparative spectral decompositions – generalize those that underlie the theoretical description of the physical world. We are using the frameworks to compare and contrast datasets recording different aspects of a single disease, such as genomic profiles of multiple cell types from the same set of patients, measured more than once by several different methods. By using the complex structure of the datasets, rather than simplifying them as is commonly done, the frameworks enable the separation of patterns of DNA alterations – which occur only in the tumor genomes – from those that occur in the genomes of normal cells in the body, and from variations caused by experimental inconsistencies. The patterns that we uncover in the data are expected to offer answers to the open question of the relation between a tumor's genome and a patient's outcome.

For example, recent comparisons of the genomes of tumor and normal cells from the same sets of ovarian and, separately, glioblastoma brain cancer patients uncovered patterns of DNA copy-number alterations that were found to be correlated with a patient's survival and response to chemotherapy. For three decades prior, the best predictor of ovarian cancer survival was the tumor's stage; more than a quarter of ovarian tumors are resistant to the platinum-based chemotherapy, the first-line treatment, yet no diagnostic existed to distinguish resistant from sensitive tumors before the treatment. For five decades prior, the best prognostic indicator of glioblastoma was the patient's age at diagnosis. The ovarian and brain cancer data were published, but the patterns remained unknown until we applied our comparative spectral decompositions.

Pending experimental revalidation, we will bring the patterns that we uncover to the clinic, to be used in personalized diagnostic and prognostic pathology laboratory tests. The tests would predict a patient's survival and response to therapy, and doctors could tailor treatment accordingly.


Physics-inspired mathematical frameworks. We develop mathematical frameworks to uncover patterns in datasets arranged in two or more third- or higher-dimensional tables, known as tensors. Rather than simplifying the datasets, as is commonly done, the frameworks make use of their complex structure in order to tease out the patterns within them.


Genomic predictors of a patient's outcome. A recent comparison of the genomes of tumor and normal cells from the same sets of ovarian serous cystadenocarcinoma patients uncovered patterns of DNA copy-number alterations that were found to be correlated with a patient's survival and response to platinum-based chemotherapy. Among patients that were diagnosed at late stages, the DNA patterns distinguished short-term survivors, with a median survival time of three years, from long-term survivors, with a median survival time almost twice as long. Among patients treated with platinum, the DNA patterns distinguished those with platinum-resistant tumors, with a median survival time of three years, from those with platinum-sensitive tumors, with a median survival time of more than seven years.


Patterns of DNA copy-number alterations in personalized diagnostic and prognostic tests. Pending experimental revalidation, we will bring the patterns that we uncover to the clinic, to be used in personalized diagnostic and prognostic pathology laboratory tests. The tests would predict a patient's survival and response to therapy, and doctors could tailor treatment accordingly. The specific genes found to be perturbed may be actively involved in cancer development and progression, and could be the basis for drug therapies.

In the News

  1. Mention: 7th most cited Proceedings of the National Academy of Sciences (PNAS) USA paper of the year 2000 and 45th most cited PNAS paper of all time, Google Scholar (February 11, 2024).
  2. Mention: BME 6770, Genomic Signal Processing course may "be pivotal for … career," Amazon Science (April 6, 2022).
  3. Mention: Among the most shared Applied Physics Letters (APL) Bioengineering research as of 2021, APL Bioengineering (October 30, 2021).
  4. Press Release: J. Kiefer, "Genome-Wide Pattern Found in Tumors from Brain Cancer Patients Predicts Life Expectancy," American Association for the Advancement of Science (AAAS) EurekAlert! (May 15, 2020).
  5. Mention: Among the top 10 most downloaded Applied Physics Letters (APL) Bioengineering articles as of 2019, APL Bioengineering (May 14, 2019).
  6. Feature: A. J. Engler and D. E. Discher, "Rationally Engineered Advances in Cancer Research," Applied Physics Letters (APL) Bioengineering 2 (3), Special Topic: Bioengineering of Cancer preface 031601 (September 2018).
  7. Mention: Among the top 10% most cited Public Library of Science (PLoS) One articles as of 2017, PLoS One (June 30, 2017).
  8. Feature: F. Pavlou, "Big Data, Hidden Knowledge," The Pathologist (June 15, 2015).4
  9. Feature: R. Atkins, "Calculating Cancer Cures," National Academy of Engineering (NAE) Innovation Podcast and Radio Series (April 19, 2015).5
  10. Press Release: J. Kiefer, "New Method Increases Accuracy of Ovarian Cancer Prognosis and Diagnosis," American Association for the Advancement of Science (AAAS) EurekAlert! (April 15, 2015).