REGISTRATION & LIGHT BREAKFAST
INTRODUCTION TO AI IN PHARMACEUTICALS
Practical Impacts of AI in Pharmaceuticals
Challenges for Delivering Machine Learning in Pharma
ARTIFICIAL INTELLIGENCE AIDED DRUG DISCOVERY
Jonathan Stokes - Broad Institute of MIT and Harvard
Machine Learning for Antibiotic Discovery
To address the antibiotic-resistance crisis, we trained a deep neural network to predict new antibiotics. We performed predictions on multiple chemical libraries and discovered a molecule from the Drug Repurposing Hub – halicin – that is structurally divergent from conventional antibiotics and displays activity against a wide spectrum of pathogens. Halicin also effectively treated Clostridioides difficile and Acinetobacter baumannii infections in mice. Deep learning approaches have utility in expanding our antibiotic arsenal.
3 Key Takeaways:
*We leveraged a message passing neural network to predict structurally novel antibiotics
*We discovered halicin, which displayed bactericidal efficacy against a broad spectrum of bacterial pathogens
*Machine learning allows us to explore vast chemical spaces in the search for new medicines
Jonathan Stokes is a Banting Fellow in the laboratory of James Collins at the Broad Institute of MIT and Harvard. He received his BHSc in 2011, graduating summa cum laude, and his PhD in antimicrobial chemical biology in 2016, both from McMaster University. His research applies a combination of chemical biology, systems biology, and machine learning approaches to develop novel antibacterial therapies with expanded capabilities over conventional antibiotics. Dr. Stokes is the recipient of numerous awards, including the Canadian Institutes of Health Research Master’s Award, the Colin James Lyne Lock Doctoral Award, and was ranked first of just 23 postdoctoral scholars to be awarded the prestigious Banting Fellowship.
Jian Tang - HEC Montreal
Graph Representation Learning for Drug Discovery
Artificial intelligence has seen big opportunities in drug discovery thanks to the large amount of data collected in this domain. Most of the data in the domain are represented as graph structures such as drug-protein interaction graph, protein-protein interaction graph, and molecular graphs. In this talk, I will introduce our recent work on developing deep learning and reinforcement learning techniques for graph representation learning and generation and their applications to drug discovery. Specifically, I will talk about (1) how to learn effective molecular graph representations for molecule properties prediction in unsupervised and supervised fashion; (2) A new generative model for molecular graph generation and optimization, which is able to generate 100% chemical valid molecules and meanwhile achieves state-of-the-art performance in both chemical properties optimization and constrained property optimization.
Pablo Cingolani - AstraZeneca
AI for Immuno Oncology on Liquid Biopsies
Immuno-Oncology tries to develop treatments and drugs that help the immune system to fight cancer. This approach has gained a lot of attention in recent years due to the success of stories of some cancer immunotherapies and checkpoint inhibitor drugs. Nevertheless, immuno-oncology does not work on every patient. We explore some challenges and opportunities when using ML and AI, particularly in the context of NGS Sequencing from ctDNA (liquid biopsies) to select biomarkers that help us understand which patient would benefit using immuno-oncology drugs.
Pablo is a leader in computational biology and life sciences with vast experience in algorithms/methods development, engineering, big data analysis, AI/machine learning and cloud/high-performance computing. At AstraZeneca, he applies bioInformatics & AI techniques to improve research and drug development in Oncology.
Automate Research Using Machine Learning Algorithms
Re-Purposing Old Drugs for New Diseases
Ola Ajayi - Biogen
Predicting Drug Interactions Using Machine Learning
Patients taking multiple medications be it prescription or over-the-counter tend to experience more adverse events. A recent survey show that simultaneous use of different products affects about half of the elderly populations with regards to increasing risk of morbidity. The goal of this solution is to develop a predictive model to determine leading indicators of drugs that are likely to interact with other drugs when administered within a specified period. The use of de-identified patient level data including demographics, prescribers, conditions and diseases as well as current and historical medications could be mined and explored to develop a novel solution. The intention is to use a mixture of publicly available data sources and real-world data in this model. Several stratifications and scenarios will be considered in the initial exploratory data analysis prior to using supervised and unsupervised learning model including ensembles to arrive at final suitable model. We will factor in groupings by similar diseases, co-morbidities, consider drug metabolism, route of administration and molecule class. A propensity or risk score will be calculated as an outcome. A data driven method of identifying drug interactions that will enhance efforts at efficacy of drugs and reduce morbidity is the objective here. The benefits include but are not limited to improving patient qualify of life, enhanced understanding of compounds, indications, contraindications for manufacturers. The final model in this solution will have a high measure of accuracy.
Using AI for Adverse Events Signalling
Adverse Effects Image Fingerprinting using CNN Deep Learning as a Means for Individual Case Safety Report Classification
DeepTox and PrOCTOR: Deep Learning in Toxicity Prediction
How AI Discovered a Drug Candidate at Deep Genomics
COMPLIANCE & ETHICS
Scalable and Accurate Machine Learning with Electronic Health Records
PANEL: The Future of AI in the Health and Pharmaceutical Industry
CONVERSATION & DRINKS
END OF DAY 1
REGISTRATION & LIGHT BREAKFAST
START UP SESSION
AI & Pharmaceutical Commercialization
Data Mining Records with Machine Learning
Programming Living Cells for Next Generation Therapeutics
PRECISION MEDICINE & PHARMACOGENOMICS
Gregory Ryslik - Celsius Therapeutics
From Single Cell Genomics to Drug Targets: Leveraging ML & AI to Discover Biological Insights
The rapid development of single cell genomics allows and the ability to understand expression level differences on a cell by cell level allows for the creation of novel precision therapies targeted toward specific subpopulations. As a tissue sample generates up to 10,000 cells and every cell can provide a read out on up to 20,000 genes, even a few hundred samples can easily generate tens of billions of data points. Identifying the signal within such large datasets requires the use of powerful machine learning algorithms throughout the entire computational biology process. In this talk we will cover the power of combining single cell genomics with machine learning in the drug discovery context and provide several examples of how we are doing this at Celsius Therapeutics. Gregory Ryslik is a statistician, data scientist and artificial intelligence researcher with experience building and leading data initiatives in companies ranging across the biotech, autotech, healthtech and fintech domains. Prior to Celsius Therapeutics, he was vice president of data science at Mindstrong Health, a healthcare company transforming mental health treatment through measurement science and artificial intelligence. Previously, Greg was the senior director and head of data science at Faraday Future, an electric vehicle startup in Los Angeles as well as the leader of the service data science group at Tesla Motors in Palo Alto. Earlier in his career, he performed machine learning research and nonclinical biostatistics research at Genentech.
Concurrently, Greg holds an adjunct assistant professor position at Pennsylvania State University and has lectured on statistics for artificial intelligence and machine learning at Stanford University Continuing Studies. He is also a fellow of the Casualty Actuarial Society, as well as a member of the American Academy of Actuaries. His research has been published in journals ranging from Nature to BMC Bioinformatics and has led to several software packages on mutational clustering.
Greg holds a Ph.D. from Yale University in biostatistics, where he researched oncogenic mutation clustering embedded within protein structure. He also holds a master’s degree in statistics from Columbia University and an undergraduate degree in mathematics, computer science and finance from Rutgers University.
Feixiong Cheng - Cleveland Clinic
Network Medicine and AI in Precision Cardio-Oncology:Learning for Cleveland Clinic Epic Systems
There are over 15.5 million cancer survivors in the United States (U.S.) alone; furthermore, cardiovascular disease is a leading cause of death and the second leading cause of mortality and morbidity in cancer survivors after recurrent malignancy in the U.S. Comorbidity between cardiovascular disease and cancer suggests an underlying shared disease etiologies, including genetic and environmental. One critical issue is that comorbidity is typically associated with various cancer treatments, termed cancer therapy-related cardiac dysfunction (CTRCD). However, there are no guidelines in terms of how to prevent and treat the new CTRCD in cancer survivors. In this talk, I will introduce a clinically relevant, network-based methodology for a comprehensive, unbiased network analysis of over 4,600 longitudinal cancer patients using clinical, laboratory and echocardiographic variables from our institutional, large-scale electronic medical records. Via network analysis, we identified four distinct subgroups who are statistically significantly correlated with incidence of CTRCD and patients’ mortality. Analysis of longitudinal patient-patient networks (20 years’ follow-up) reveals dosing-time-dependent (‘chronopharmacology’) CTRCD. Using clinical variable network analysis, we identified several clinically relevant predictors (i.e., Troponin-T and NT-proBNP) that are significantly associated with patients’ mortality. Compared to traditional machine learning approaches, network methodologies are more interpretable, visualizing the clinical decision boundary of cancer patients with CTRCD.
Feixiong Cheng, PhD, is a principal investigator with Cleveland Clinic’s Genomic Medicine Institute. Dr. Cheng is a computational and systems biologist by training, with expertise in analyzing, visualizing, and mining data from real world (e.g., electronic health records, and health care claims) and experiments that profile the molecular state of human cells and tissues by interactomics, transcriptomics, genomics, proteomics, and metabolomics for precision medicine drug discovery and patient care. Dr. Cheng is working to develop computational and experimental network medicine technologies for advancing the characterization of disease heterogeneity, thereby approaching the goal of coordinated, patient-centered strategies to innovative diagnostics and therapeutics development. The primary goal of Dr. Cheng’s lab is to combine tools from genomics, network medicine, bioinformatics, computational biology, chemical biology, and experimental pharmacology and systems biology assays (e.g., single cell sequencing and iPS-derived cardiomyocytes), to address the challenging questions toward understanding of various human complex diseases (e.g., cardio-oncology, pulmonary vascular diseases, and Alzheimer’s disease), which could have a major impact in identifying novel real-world data-driven diagnostic biomarkers and therapeutic targets for precision medicine. From 2013 to 2017, Dr. Cheng was trained as Postdoctoral Research Fellow in the field of pharmacogenomics and network medicine across Vanderbilt University Medical Center, Northeastern University, and Dana-Farber Cancer Institute. During 2017-2018, Dr. Cheng was promoted to Research Assistant Professor working with two of the world’s leading experts in the field of network medicine, Drs. Albert-Laszlo Barabasi and Joseph Loscalzo, with dual appointment at Northeastern University and Harvard Medical School. Dr. Cheng has received several awards, including NIH Pathway to Independence Award (K99/R00), SCI highly cited papers reward, and Vanderbilt Postdoc of the Year Honorable mention.
Niha Beig - Case Western Reserve University
Keeping Gender in Mind: Applications of Machine Learning in Neuro-Oncology
Investigation of the sexually dimorphic expression of genes has reported that the molecular-level effects of gender influence treatment response and prognosis in various cancers. Hence, there is a need to develop “gender-specific” models that are prognostic of patient outcome to potentially assist in building comprehensive and patient-centric treatment plans. In the field of neuro-oncology, robust and consistent evidence shows that males are almost twice as likely to develop Glioblastoma (GBM), a rapidly fatal primary brain tumor, compared to females (1.5:1). Even though the visual appearance of different male and female tumor phenotypes on MRI look similar, there nonetheless might be subtle sub-visual cues reflective of the differences in the micro-architectural appearance, that are not visible to the naked human eye. Radiomics can provide a surrogate mechanism to non-invasively characterize these GBM tumor by capturing sub-visual cues of morphologic diversity (e.g. roughness, image homogeneity, regularity and edges) on routine MRI scans. It is also critical to identify the molecular associations of these radiomic features in gender controlled cohorts with underlying signalling pathways that drive different biological processes, via radiogenomic analysis. Such cross-scale associations using radiomics and radiogenomics in GBM could allow for designing gender-specific personalized treatments.
Niha Beig is currently a PhD researcher in Brain Image Computing Laboratory at Case Western Reserve University, Cleveland Ohio, USA. Her research focus is on developing assistive diagnostic tools in the field of neuro-oncology using data science and machine learning techniques. She has been working on drawing insights from quantitative medical images and genomic sequences to help distinguish various subtypes of cancers in adult and pediatric gliomas. She has over 30 peer-reviewed abstracts and publications in the field of radiomics and radiogenomics.
Niha is currently an elected member of the Sigma Xi: Scientific Research Honor Society and has also served as an area chair for Women in Machine Learning Workshop in 2018 and 2019.
Artificial Intelligence & Pharmacogenomics
AI and Machine Learning Approaches in Clinical Trials
Machine Learning to Effectively Match Patients and Trials
Unleashing the Power of your Data
Boost Clinical Performance by Decreasing Data Entry Errors
PANEL: Is Inter-industry Collaboration Essential for the Successful Integration of AI in Pharmaceuticals?
END OF SUMMIT