Samuel D. Conte Distinguished Lecture Series

Professor Carl Kingsford

Title

Automatically configuring bioinformatics tools in genomics and pan-genomics

Abstract

Measurements from genomic sequencing experiments (such as RNA-seq and Hi-C) must be transformed to be useful for revealing biological insights. This requires using software that is often highly parameterized and complex, leading to time-consuming and error-prone manual decisions about computational analysis parameters. In this talk, I will present two efforts to automate the analysis decisions that must be made when processing sequencing data.

In the first part, I will present a method for selecting an appropriate reference genome from a pan-genome for the analysis of genomic 3D structure measurements (Hi-C experiments). This is particularly important when studying samples with large-scale genomic variants (e.g. cancer samples). Our approach combines pan-genomic data structures with algorithms to infer optimal sample-specific reference genomes. I will show that this leads to better identification of topologically associating domains, an important element of genomic 3D structure.

In the second part, I will present a method for automatically learning parameters for blackbox analysis steps in bioinformatics pipelines. I will describe a framework that integrates Bayesian optimization and contrastive learning to train a predictor for optimal, sample-specific, parameters for bioinformatics analysis programs. Using the example of transcript assembly, I will show that the framework often leads to significant gains in assembly accuracy.

These efforts enable more accurate, more reproducible, lower-cost computational workflows to extract biological insight from large-scale data sets.

Joint work with Yihang Shen, Zhiwen Yan, Yutong Qiu, Lingge Yu, and Tianyu Zhang.

Bio

Carl Kingsford is the Herbert A. Simon Professor of Computer Science in the Ray and Stephanie Lane Computational Biology Department at Carnegie Mellon University. His research focuses on developing new computational methods for analysis of large biological data sets, especially genomic and transcriptomic data. He is the Director of CMU's Center for Machine Learning and Health, Co-Director of CMU’s Ph.D. program in computational biology, and Co-Founder of Ocean Genomics, Inc. He earned a Ph.D. in Computer Science from Princeton University in 2005 and was named a Fellow of the ISCB in 2024.

Professor Robin Murphy

Title:

Robots (and Research) to the Rescue

Abstract:

Ground, aerial, and marine robots are increasingly used by responders to save lives, mitigate ongoing threats, and accelerate economic recovery. The recent Surfside condo collapse and Hurricane Ian are two examples of the extreme environments that robots, and their operators, must function in. Clearly, rescue robots have great societal benefit; however our work at these two disasters illustrate why disaster robotics is important to robotics research in general. One reason is that research in the field at a disaster informs the virtuous research cycle, guiding both fundamental and convergent research. The use of robots at a disaster provides a “canary in the coal mine” indication of gaps in hardware, software, and human-robot interaction that might take years to discover through hypothesis-driven laboratory testing. The use of drones at the Surfside collapse has led to fundamental research in reconstructing voids using photogrammetry.  Hurricane Ian showed that drone pilots by the end of the second day of the response were showing fatigue and cognitive deficits equivalent to being legally drunk in most states. Surprisingly, the fatigue did not lead to aviation errors, possibly because of the robustness and automation of the drones being used, but squads made significant errors in collecting the imagery needed by incident command. Hurricane Ian has produced advances in computer vision and machine learning, including new schemas for coding, identifying alignment errors, and producing a massive labeled open-source dataset. A second reason why disaster robotics is valuable is that it, by necessity, is pioneering domain-inspired, interdisciplinary synthesis, which in turns calls for new pedagogical approaches for educating the next generation of scientists.

Bio:

Robin R. Murphy, Ph.D. (’92) and M.S. (‘89) in computer science and B.M.E. (‘80) from the Georgia Institute of Technology,  is the Raytheon Professor of Computer Science and Engineering at Texas A&M University and a director of the Center for Robot-Assisted Search and Rescue. Her research focuses on artificial intelligence, robotics, and human-robot interaction for emergency management. She is an AAAS, ACM, and IEEE Fellow, a TED speaker, and author of over 400 papers and four books including the award-winning Disaster Robotics which captures much of her research deploying ground, aerial, and marine robots to over 30 disasters in five countries including the 9/11 World Trade Center, Fukushima, Hurricane Harvey, and the Surfside collapse. Her contributions to robotics have been recognized with the ACM Eugene L. Lawler Award for Humanitarian Contributions and a US Air Force Exemplary Civilian Service Award medal.  Dr. Murphy has served on numerous professional and government boards, including the Defense Science Board and National Science Foundation, as well as the AI for the Benefit of Humanity prize committee.

Professor Somesh Jha

Title

Trustworthy Machine Learning and the Security Mindset

Abstract

Fueled by massive amounts of data, models produced by machine-learning (ML) algorithms, especially deep neural networks (DNNs), are being used in diverse domains where trustworthiness is a concern, including automotive systems, finance, healthcare, natural language processing, and malware detection. Of particular concern is the use of ML algorithms in cyber-physical systems (CPS), such as self-driving cars and aviation, where an adversary can cause serious consequences. Interest in this area of research has simply exploded. In this work, we will emphasize the need for a security mindset in trustworthy machine learning, and then cover some lessons learned. Large Language Models (LLMs) as been a paradigm shift and towards the end we will touch on the subject of trustworthiness in the context of LLMs.

Bio

Somesh Jha received his B.Tech from Indian Institute of Technology, New Delhi in Electrical Engineering. He received his Ph.D. in Computer Science from Carnegie Mellon University under the supervision of Prof. Edmund Clarke (a Turing award winner). Currently, Somesh Jha is the Lubar Professor in the Computer Sciences Department at the University of Wisconsin (Madison). His work focuses on analysis of security protocols, survivability analysis, intrusion detection, formal methods for security, and analyzing malicious code. Recently, he has focused his interest on privacy and adversarial ML (AML). Somesh Jha has published several articles in highly-refereed conferences and prominent journals. He has won numerous best-paper and distinguished-paper awards. Prof. Jha received the CAV award for his work on CEGAR and also has received the IIT-Delhi Distinguished Alumni award. Prof. Jha is the fellow of the ACM, IEEE, and AAAS.