ai and machine learning Archives | LSP
Did AI Solve the Protein-Folding Problem?
LSP alumni Nazim Bouatta and Mohammed AlQuraishi were featured in Harvard Magazine last month, sharing their perspectives on the next phase for AI in scientific discovery. Although some headlines suggest that AI has “solved” the protein folding problem, Bouatta and AlQuraishi believe that many larger challenges remain. They hope tools like AlphaFold and OpenFold will shed light on protein dynamics and clarify how proteins interact functionally with other proteins or therapeutic agents.
A Step Toward New Solutions for Inflammatory Pain
A new study from Clifford Woolf and Peter Sorger revealed novel, bi-directional communication between pain-sensing neurons and immune cells, with different inflammatory stimuli leading to unique cellular responses. These findings increase our understanding of the complex interactions that control pain and bring us closer to developing more targeted analgesics.
A New Tool for Diagnosing Cancer
New results published in Nature Cancer detail a proof-of-principle study of Orion, a new imaging tool developed by a team at the LSP and RareCyte, Inc. Orion serves as a bridge between the histology, the standard-of-care method for diagnosing tumors in the clinic, and high-plex imaging methods that provide rich molecular details about cell types and states. Orion could help usher in a new era for pathology, where pathologists use Orion in the clinic to inform disease diagnosis and treatment.
Benjamin Gyori selected for the DARPA ASKEM program
By Juliann Tefft | November 4, 2022
Benjamin M. Gyori, PhD, Director of the Machine Assisted Modeling & Analysis Platform at the Lab of Systems Pharmacology at Harvard Medical School, received a 3.5-year research grant from the Defense Advanced Research Projects Agency (DARPA) as part of the Automating Scientific Knowledge Extraction and Modeling (ASKEM) program.
The ASKEM program aims to leverage artificial intelligence (AI) approaches to accelerate scientific modeling. Models of complex systems have become an increasingly critical part of daily life; from predicting the weather to deciding which drug will best treat a particular disease, models help people predict outcomes and make informed decisions. However, models are laborious to build and maintain in the face of rapidly evolving information, and therefore become quickly obsolete and difficult to reuse. As an example, over the course of the COVID-19 pandemic, the scientific understanding of how the virus spreads – from surfaces, human contact, or air – changed rapidly, and numerous variants of the virus emerged with evolving characteristics. We can imagine that incorporating the most up-to-date data and knowledge into predictive models could critically impact how health policymakers advise the public to stay safe. The DARPA ASKEM program hopes to help address this broad problem by developing new technology to create a machine-assisted modeling framework that can be applied to multiple scientific areas, including viral pandemics and space weather.
Gyori will work with LSP members Charles Tapley Hoyt, PhD, and Klas Karis, MS to develop the Modeling with an Intelligent Research Assistant (MIRA) system. MIRA aims to make the process of generating, reusing, and updating scientific models more efficient through two main efforts: creating a template-based meta-modeling framework and technology for rapidly assembling domain knowledge graphs that can facilitate modeling. When experts develop a model within their specialized domain, they use their high-level understanding to encode assumptions into the mathematical equations within their model. These assumptions are often non-obvious to external users, thus inhibiting both model reuse and updates.
MIRA’s meta-model templates will capture the individual concepts, processes, and assumptions that make up a given model, making it possible to intuitively compare and repurpose existing models. MIRA will also assemble the ever-evolving scientific knowledge into domain knowledge graphs (DKG), which link models to findings in underlying scientific literature as well as structured ontologies. These efforts build on Gyori’s ongoing expertise in text-mining for large-scale knowledge assembly using the Integrated Network and Dynamical Reasoning Assembler (INDRA). Gyori’s team’s development of MIRA will help streamline the process of generating scientifically grounded mathematical models and make it possible to extend existing models to new territories. Together, these efforts could significantly improve our ability to predict and respond to future events like pandemics.
Congratulations, Ben – we look forward to seeing what you accomplish through this exciting program!
To find out more about Ben’s team’s work, see https://indralab.github.io/.
Ben Gyori Awarded a DARPA Director’s Fellowship
By Juliann Tefft | July 8, 2022
Benjamin M. Gyori, Ph.D., Director of the Machine Assisted Modeling & Analysis Platform at the Lab of Systems Pharmacology at Harvard Medical School, received the Director’s Fellowship Award from the Defense Advanced Research Projects Agency (DARPA). This prestigious award extends Gyori’s 2020 DARPA Young Faculty Award (YFA) for an additional year and further recognizes Gyori’s impressive accomplishments as an early career principal investigator.
Gyori’s work focuses on how artificial intelligence can help humans interact with biological knowledge. A massive problem in biomedicine is that the rate of publications outpaces what human scientists can manage to read, which generates an intractable amount of new data and knowledge each year. To help, we need new and better ways to monitor new knowledge as it appears and integrate these discoveries with prior knowledge in an actionable form. That’s where Gyori and the INDRA team come in.
Our project is a step towards solving this problem. Our work can be applied to many other domains as well – we aim to help any field where information around a complex, interconnected system is rapidly evolving, and people need to make informed decisions in the face of evolving information. – Ben Gyori
Information overload has been particularly acute during the COVID-19 pandemic. Three hundred new COVID-19-related papers are published each day – as of July 2022, there are more than 250,000 papers about COVID on PubMed. In theory, each paper contains data that can inform important public health and therapeutic decisions, but filtering through the information is a considerable challenge in the face of such a large body of work.
With the DARPA Young Faculty Award, Gyori and his team members Charles Hoyt and Klas Karis used machine-reading and scalable knowledge assembly algorithms to create a knowledge graph database from the biomedical literature and other structured data sources. The graph database contains over 300 million relationships, which represent various biological information – like causal mechanisms, experimental findings, and properties of proteins, small molecules, diseases, and clinical trials – all linked to primary evidence in the literature. Their graph database supports both machine-assisted analysis of experimental data and human-machine dialogue (i.e., interaction through the English language), to allow researchers to make discoveries using this knowledge.
With the additional Director’s Fellowship Award funding, Gyori and his team will develop a public web portal that will allow users to analyze and explore the data within the automatically assembled knowledge graphs. They also plan to pursue several projects to apply these rich biomedical datasets, such as to infer drug mechanisms of action and explain unexpected drug side effects.
This award is a huge honor that recognizes the important work we’ve been doing over these past few years. It shows that our technology has the potential for many exciting applications – within biomedicine and beyond. I’m thrilled to continue expanding on our research to understand and manage more complex systems. – Ben Gyori
To find out more, visit INDRA labs or follow us on Twitter at IndraSysBio.
Ben Gyori named 2018 DARPA Riser
By Laura Maliszewski | September 6, 2018
Benjamin M. Gyori, Ph.D., A Research Scientist in Therapeutic Science was selected as one of 50 DARPA Risers: up and coming early career investigators with the potential to perform innovative research at the frontiers of science and technology relevant to the mission of DARPA.
As a DARPA Riser, Ben will attend DARPA’s 60th Anniversary Symposium (D60) at Gaylord National Harbor, Sept 5-7, and present his proposal on a computer system which autonomously monitors events and scientific discoveries, integrates them into actionable models, and proactively reports relevant analysis results.
D60 will bring together 1,500 forward-thinking scientists, engineers, and other innovators interested in sharing ideas and learning how DARPA has shaped and continues to shape breakthrough technologies. D60 will also feature Dr. Peter Sorger, who leads the Harvard Program in Therapeutic Science, as an invited speaker on the Accelerating Science panel where he will present on AI technologies for bridging mathematical models and cellular biology.
Ben’s current work focuses on computational approaches to accelerate scientific discovery. He has been an active performer in the DARPA Big Mechanism, Communicating with Computers, World Modelers, and Automated Scientific Discovery Framework programs.
Gyori is a co-developer of INDRA, a system which automatically assembles information about biochemical mechanisms extracted from the scientific literature into explanatory and predictive models. He is also leading the development a collaborative dialogue system allowing a human user to talk with a machine partner to learn about molecular mechanisms, and formulate, and test model hypotheses.