Abstracts for Round Table (RT) Session 1: Artificial Intelligence Applications in Pharmaceuticals
Moderators: Lina Niu and Sandip Tiwari
Presentation 1
Operation Oppenheimer: how we get in and out of the Machine Learning (ML) modeling rabbit hole for LC retention prediction
Jane Kawakami, PhD, Principal Scientist, Pfizer Inc.
Abstract:
When a new impurity arises due to synthetic route changes or process adjustments, a lab scientist will typically observe it first in a chromatogram when analyzing the samples, often by a LC method. Based on our current workflow, to ensure this impurity does not affect drug stability and safety, an analyst will consult with a mass spectrometry (MS) spectroscopist to determine the potential 2D structures and then confirm it via NMR analysis, which requires pure, isolated materials. Depending on the complexity, the synthesis of a compound could take weeks and cost $20-30k dollars. Occasionally, it may not even be feasible to carry out the synthesis due to reactivity/stability issues.
Integrating MS and NMR with QSRR (Quantitative Structure Retention Relationship) modeling offers a promising approach to efficiently narrow down the potential impurity structures, resulting in reduced time and cost for structural identification. Our work aims to improve accuracy and increase automation in QSRR modeling by exploring various descriptors and modifying existing ML workflows. We herein demonstrated the feasibility of (1) developing local QSRR models for RPLC or HILIC based RSM/API/DP assay and purity methods across multiple projects and (2) leveraging the models to support new impurity identification. Despite the relatively small training dataset (fewer than 25 compounds), we observed that by exploring molecular descriptors and ML algorithms robust models can be achieved.
Bio:
Jane Kawakami is a Principal Scientist in Analytical Research and Development under Pharmaceutical Sciences of Small Molecules at Pfizer. In her current role, she leads a method development group that focuses on utilizing LC-UV-MS (Liquid Chromatography-UV-Mass Spectrometry) for small molecule drug characterization and developing separation-related technologies, including multidimensional separations and chromatographic retention prediction via both mechanistic and data-driven modeling.
Her research experiences prior to Pfizer between 2014 and 2021 include development of miniaturized LC instrumentation, biomarker discovery for Alzheimer’s disease using LC-MS based non-targeted lipidomics, as well as development of microfluidic immunoassays for biomarker research. Outside of work, she enjoys running, dancing and swimming.
Presentation 2
Management of AI technologies
Jordan W. Suchow, PhD, Assistant Professor, Stevens Institute of Technology
Abstract:
Artificial intelligence can no longer be discussed as something that is on the horizon. Intelligent machines already play a role in many industries and their responsibilities will only continue to grow. This does not mean the end for humans in the workplace — but it does involve a new set of skills and a broader perspective on the kinds of challenges that are starting to confront leaders in the management of AI technologies. In this talk, I will discuss management of AI technologies in the context of pharmaceuticals, including a focus on advances in privacy-preserving AI, applications to scientific due diligence, and the integration of AI into the scientific process itself.
Bio:
Jordan W. Suchow is an assistant professor of Information Systems at Stevens Institute of Technology, where he performs research at the intersection of cognitive science and information systems. Suchow earned a B.S. in computer science from Brandeis and A.M. and Ph.D. from Harvard in cognitive psychology before performing postdoctoral research at UC Berkeley and serving as a PI on the DARPA NGS2 program. At Stevens, Suchow teaches courses on the management of AI technologies, which he developed as part of a certificate program in business intelligence and analytics on the same.
Presentation 3
A highly accurate, reliable, and efficient CSP platform and FEP+ solubility prediction workflow for small molecule drug formulation
Shiva Shekharan, PhD, Director, Business Development, Schrodinger
Abstract:
Early assessment of crystal polymorphism and thermodynamic solubility continues to be elusive for drug discovery and development despite its critical importance especially for the ever-increasing fraction of poorly soluble drug candidates. We have developed a crystal structure prediction (CSP) method that combines a novel systematic crystal packing search algorithm and a hierarchical energy ranking protocol to predict crystal polymorphs. The predictive power of our CSP is validated on a large and diverse dataset of 65 druglike molecules with 135 experimentally found polymorphic forms to complement experimental polymorph screening workflows. The physics-based free energy perturbation (FEP+) approach for computing thermodynamic aqueous solubility is assessed across a diverse chemical space spanning several pharmaceutically relevant compounds in the literature. The high accuracy, reliability, and efficiency of our CSP and FEP+ methods with large scale validations is designed to support polymorph screening and solvent screening experiments in drug substance and drug product development processes.
Bio:
Shiva Sekharan is the Senior Director of Formulations Business Development at Schrodinger, Inc. and is responsible for driving the business development efforts in the formulations space. Shiva is an experienced business development executive in the CRO and AI-based services and software solutions industry and has several years of experience in managing business accounts, customer relationships, and expectations with clients in the pharmaceutical, agrichemical and academic industries across the US, Europe and Asia territories. His expertise lies in identifying new business opportunities among existing customers, devising sales and collaboration strategies for customer expansion, ensure top-tier services, products and knowledge-driven solutions are available 24/7 to customers across the globe.
Before arriving at Schrodinger, Shiva held a BD role at XtalPi Inc., where he led the US solid-state services unit, worked with departmental heads to establish effective goals, sales targets, outline procedures and best practices and provided strategic directions to increase revenue.
Shiva earned his Ph.D. in Theoretical Chemistry from the University of Duisburg-Essen in Germany followed by postdoctoral stints at the Max-Planck Institute for Polymer Science, Emory University, Fukul Institute for Fundamental Chemistry and Yale University. Shiva is an accomplished computational chemist with strong research expertise in the areas of quantum chemistry and drug discovery (>40 publications, >1100 citations, H-index = 20).
Moderators: Lina Niu and Sandip Tiwari
Presentation 1
Operation Oppenheimer: how we get in and out of the Machine Learning (ML) modeling rabbit hole for LC retention prediction
Jane Kawakami, PhD, Principal Scientist, Pfizer Inc.
Abstract:
When a new impurity arises due to synthetic route changes or process adjustments, a lab scientist will typically observe it first in a chromatogram when analyzing the samples, often by a LC method. Based on our current workflow, to ensure this impurity does not affect drug stability and safety, an analyst will consult with a mass spectrometry (MS) spectroscopist to determine the potential 2D structures and then confirm it via NMR analysis, which requires pure, isolated materials. Depending on the complexity, the synthesis of a compound could take weeks and cost $20-30k dollars. Occasionally, it may not even be feasible to carry out the synthesis due to reactivity/stability issues.
Integrating MS and NMR with QSRR (Quantitative Structure Retention Relationship) modeling offers a promising approach to efficiently narrow down the potential impurity structures, resulting in reduced time and cost for structural identification. Our work aims to improve accuracy and increase automation in QSRR modeling by exploring various descriptors and modifying existing ML workflows. We herein demonstrated the feasibility of (1) developing local QSRR models for RPLC or HILIC based RSM/API/DP assay and purity methods across multiple projects and (2) leveraging the models to support new impurity identification. Despite the relatively small training dataset (fewer than 25 compounds), we observed that by exploring molecular descriptors and ML algorithms robust models can be achieved.
Bio:
Jane Kawakami is a Principal Scientist in Analytical Research and Development under Pharmaceutical Sciences of Small Molecules at Pfizer. In her current role, she leads a method development group that focuses on utilizing LC-UV-MS (Liquid Chromatography-UV-Mass Spectrometry) for small molecule drug characterization and developing separation-related technologies, including multidimensional separations and chromatographic retention prediction via both mechanistic and data-driven modeling.
Her research experiences prior to Pfizer between 2014 and 2021 include development of miniaturized LC instrumentation, biomarker discovery for Alzheimer’s disease using LC-MS based non-targeted lipidomics, as well as development of microfluidic immunoassays for biomarker research. Outside of work, she enjoys running, dancing and swimming.
Presentation 2
Management of AI technologies
Jordan W. Suchow, PhD, Assistant Professor, Stevens Institute of Technology
Abstract:
Artificial intelligence can no longer be discussed as something that is on the horizon. Intelligent machines already play a role in many industries and their responsibilities will only continue to grow. This does not mean the end for humans in the workplace — but it does involve a new set of skills and a broader perspective on the kinds of challenges that are starting to confront leaders in the management of AI technologies. In this talk, I will discuss management of AI technologies in the context of pharmaceuticals, including a focus on advances in privacy-preserving AI, applications to scientific due diligence, and the integration of AI into the scientific process itself.
Bio:
Jordan W. Suchow is an assistant professor of Information Systems at Stevens Institute of Technology, where he performs research at the intersection of cognitive science and information systems. Suchow earned a B.S. in computer science from Brandeis and A.M. and Ph.D. from Harvard in cognitive psychology before performing postdoctoral research at UC Berkeley and serving as a PI on the DARPA NGS2 program. At Stevens, Suchow teaches courses on the management of AI technologies, which he developed as part of a certificate program in business intelligence and analytics on the same.
Presentation 3
A highly accurate, reliable, and efficient CSP platform and FEP+ solubility prediction workflow for small molecule drug formulation
Shiva Shekharan, PhD, Director, Business Development, Schrodinger
Abstract:
Early assessment of crystal polymorphism and thermodynamic solubility continues to be elusive for drug discovery and development despite its critical importance especially for the ever-increasing fraction of poorly soluble drug candidates. We have developed a crystal structure prediction (CSP) method that combines a novel systematic crystal packing search algorithm and a hierarchical energy ranking protocol to predict crystal polymorphs. The predictive power of our CSP is validated on a large and diverse dataset of 65 druglike molecules with 135 experimentally found polymorphic forms to complement experimental polymorph screening workflows. The physics-based free energy perturbation (FEP+) approach for computing thermodynamic aqueous solubility is assessed across a diverse chemical space spanning several pharmaceutically relevant compounds in the literature. The high accuracy, reliability, and efficiency of our CSP and FEP+ methods with large scale validations is designed to support polymorph screening and solvent screening experiments in drug substance and drug product development processes.
Bio:
Shiva Sekharan is the Senior Director of Formulations Business Development at Schrodinger, Inc. and is responsible for driving the business development efforts in the formulations space. Shiva is an experienced business development executive in the CRO and AI-based services and software solutions industry and has several years of experience in managing business accounts, customer relationships, and expectations with clients in the pharmaceutical, agrichemical and academic industries across the US, Europe and Asia territories. His expertise lies in identifying new business opportunities among existing customers, devising sales and collaboration strategies for customer expansion, ensure top-tier services, products and knowledge-driven solutions are available 24/7 to customers across the globe.
Before arriving at Schrodinger, Shiva held a BD role at XtalPi Inc., where he led the US solid-state services unit, worked with departmental heads to establish effective goals, sales targets, outline procedures and best practices and provided strategic directions to increase revenue.
Shiva earned his Ph.D. in Theoretical Chemistry from the University of Duisburg-Essen in Germany followed by postdoctoral stints at the Max-Planck Institute for Polymer Science, Emory University, Fukul Institute for Fundamental Chemistry and Yale University. Shiva is an accomplished computational chemist with strong research expertise in the areas of quantum chemistry and drug discovery (>40 publications, >1100 citations, H-index = 20).
