Leaders in Pharmaceutical Business Intelligence (LPBI) Group

eProceeding 2019 Koch Institute Symposium – 18th Annual Cancer Research Symposium – Machine Learning and Cancer, June 14, 2019, 8:00 AM-5:00 PM ET MIT Kresge Auditorium, 48 Massachusetts Ave, Cambridge, MA

Announcement

Aviva Lev-Ari, PhD, RN,

Founder and Director of LPBI Group will be in attendance covering the event in REAL TIME

@pharma_BI

@AVIVA1950

#KIsymposium

AGENDA

https://custom.cvent.com/C09E632203614BC295DA40B01F45218F/files/47a27352fc1f4eb486236a34203f6396.pdf

https://ki.mit.edu/news/symposium/2019

 

8:00–9:00 Coffee Break Registration

 

9:00 –9:10 Introductory Remarks Tyler Jacks Koch Institute, MIT Phillip Sharp Koch Institute, MIT

• INTERDISCIPLINARY research at the interface of Cancer with other disciplines, this year CS, ML,

9:10–10:30 Machine Learning in Cancer Research: The Need and the Opportunity

Moderator: Phillip Sharp, Nobel Laureate 

• ML and cancer @MIT over a decade under discussion, ML & Cancer in research is embryonic
• REENGINEERING Healthcare using ML: understand disease, treating it
• MIT Stephen A. Schwarzman College of Computing, AI and Healthcare – an MIT commitment of a $MILLION
• COST OF DRUGS IS A THIRD IN TREATMENT — TOO HIGH,

James P. Allison MD Anderson Cancer Center, Nobel Laureate FOR CURE OF CANCER, THE ONLY ONE

Immune checkpoint blockade in cancer therapy: Historical perspective, new opportunities, and prospects for cures

• long survival on ipilimumab – Lung cancer pleural effusion
• CTLA-4Blockade – attenuated or Terminate Proliferation vs unrestrained Proliferation – APC
• PD-1 – Ipi/Nivo vs. Ipi in metastatic Melanoma – 10 years survival after initial three years
• FDA Approved CTLA-4 and Anti-PD-1 for a dozen indication
• CD4, CD8, NKT – CHeckpoint blockage modulates MC38 infiltrating T cell population frequency
• Cellular mechanism interaction
• T cell differentiation is complex – negative costimulation regulate T cell differentiation
• comprehensive profiling of peripheral T cell arising
• Expansion of phynotype – distinct gene expression
• Transcription loss of PD-1 expands CD8 T cells
• Nuanced Model – T cell differentiation
• Role of T cells in Immunotherapy

Aviv Regev Broad Institute; Koch Institute, MIT

Cell atlases as roadmaps to cancer

• Experimental data and computation data
• Tumors: a complex cellular ecosystem
• Genetic pathology, Combination therapy Genetic variants and
• Components vs Programs (regulatory, pathways, tissue modules, disease road map)
• ML Applications: Immunotherapy in melanoma
• Learning an single cell profiling and spatial organization – massively parallel cell profiling
• malignant cells from T cell cold tumors
• T cell T cell CD8 Malignant: Resistant
• Computational search predicts CDK4/6 as program regulators
• the resistance program is reversed abemaciclib sensitized to B16
• Resistance to immunotherapy in Melanoma
• Generalize across types of cancers – Resistance to treatment
• Atlas for Synovial sarcoma has anti-tumor activity in its immune environment
• Melanoma vs Synovial sarcoma (SS18-SSX fusion directly controls the core oncogenic program
• Target the core oncogenic program therapeutically
• IFN from CD8 T NF repress the program
• Insi2Vec: Define a cell by intrinsic and extrinsic and spatial features: generalize across patient samples in melanoma
• Identifying key patio-molecular archetypes
• Expanding to full Transcriptome
• Emerging toolbox for spatial genomics: Protein, RNA,
• Compressed sensing,

Regina Barzilay MIT Computer Science and Artificial Intelligence Laboratory; Koch Institute,

MIT How machine learning changes cancer research

• ML & Clinical Oncology
• Data Science perspective – Raw Data to Cure: Extract information
• NLP – penetration of deepLearning @MGH
• Drug discovery – New molecules
• MIT – ML in drug discovery
• Risk models: age, family history, prior breast procedures breast density
• High risk and Breast density
• Myth #1: DEMOCRATIZATION OF AI – deep learning and customization
• Myth @2: Bias – Deep Learning allows across population studies
• Myth #3: Black-Box Architecture 0 @MIT Interpretable models
• J-Clinic @MIT – ML and Healthcare Hospitals are recruited to use the algorithms

11:00–12:25 Machine Learning to Decipher Cellular and Molecular Mechanisms in Cancer

Moderator: Matthew Vander

Heiden Michael Angelo Stanford University

Comprehensive enumeration of immune cells in solid tumors using Multiplexed ion beam imaging (MIBI)

• Tissue imaging – high dimensional, multiplexing: high sensitivity, dynamic range,
• GI Tissue selection – Immunofluorescence (IF)  fluorophores – elemental reporters and mass spectrometry  Label antibody,
• Immune populations in triple negative breast cancer
• Patient stratification survival
• mixed tumor: Immune cells, non immune cells – response to PD-1
• Single cell: imaging data trained by nuclear interior ( nuclei only, , segmentation channel using DeepCell v1.0
• TB granulomas,
• Cell type identification with deep learning – No downstream
• 3D Segmentation Melanoma
• Assays and Ion sources

Olga Troyanskaya Princeton University

Decoding cancer genomes with deep learning

• genome to phenotype – BIG genomics data to understand human disease
• exosome – non coding region
• gene regulation affected by genome for gene expression
• sequence specific
• Interpreting a genome: predicting the effect of any mutation: coding variant vs non-coding variance – regulatory code, transcription marks, DeepSEA – model learns from one genome interactions of biochemical level models, variants motives cell type
• Interpreting a Genome: predicting the effect of any mutation: Noncoding variant effect in ASD from WGS data
• DeNovo mutations: How impactful are these mutations in Autism patients – to % not explained by sequencing
• DeepSEA identifies significant noncoding regulatory mutation burden in ASD
• Cell-type specific expression models capture tissue-specific
• ExPecto-prioritized putative casual GWAS variants
• HumanBase Networks best model selected are

Dana Pe’er Sloan Kettering Institute

Manifolds underlying plasticity in development, regeneration and cancer

• Asynchronous Nature of Cells
• Single cell transitional states, homeostatic
• Representing population structure as a neighbor graph: each node is a cell connected to a neighborhood – diffusion components Magic: Using structure for data recovery
• PALANTIR: a probabilistic model for cell fate
• Learn differentiation trajectories/pseudo-time ordering
• Computing Differentiation probabilities stochastic process – branch
• The gut tube comprises cells of two distinct regions
• mammal cell plasticity – endodermal organ identities in space
• VE cells express key organ lineage regulators similarly to DE cell
• Uncovering unappreciated lineage relationship
• Lung adenocarcinoma – tumor – 10% are cancer cell the rest are not therefore – they are ORGANS
• Primary tumor exhibit epithelial progenitor cells
• Pancreatic Ductal Adenocarcinoma – ductal Metaplasia
• Immune ecosystem reprogrammed
• Leptomeningeal niche — iron binding – adopt environment and prevent iron from the immune system
• Tumor plasticity – Mutation Kras + injury to cause pancreatic cancer

 

Peter Sorger Harvard Medical School

Measuring and modeling variability in drug response in cells, tissues and clinical trials

• combination therapy and drug synergy
• Re-digitalization of 300 clinical trials
• combination immunotherapy: Assumption each patient benefits from both = synergy
• vs the assumption of NO interaction: each Patient benefit from ONE or the Other
• Novartis PDX models
• Gastric Cancer PDXs
• Three way combination drug therapy – a deeper look at RAF and MEK inhibitors in BRAF mutant melanoma
• Pharmacology of ‘reconstituted pulses” independent action
• Heterogeneous response – sophisticated stratification of patients Four biopsy vs Three Biopsy
• Highly multiplexed, high resolution image of FFPE specimens
• Assaying the effects of multiple possible
• Cancer Precision Medicine per NCI
• Precision medicine based on therapeutic response at sisngel cell level
• Pharmacological synergy is over emphsized – time sensitivity prevent the interactions expected

12:30–2:00 Lunch Break

2:00–3:00 Big Data, Computation and the Future of Health Care

Moderator: Susan Hockfield

Jay Bradner Novartis Institutes for BioMedical Research

• AI is a Tool in all phases of drug discovery integrated analysis of epigenetics
• strategy for investment in 400 Scientists, big data
• monitoring of Clinical Trial around the  World
• generative chemistry – 100 years of discovery in chemistry
• validation models – bring trials to patients – distributed model of participation
• catalyst – ML in biological experimentation
• Published 700 articles per year at Novartis by 400 scientists
• Population Data sets
• IN 5 YEARS MASSIVE data will dominate molecule selection in Drug delivery

Aine Hanly VP Amgen

• 17% improvement in survival rate in 50 years
• Data infrastructure investment paid off
• Partnerships, interoperability in EMR

Clifford Hudis American Society of Clinical Oncology (ASCO)

• 75% of communication is by FAX still
• White cell data count – convert data structure on EMR
• Data in the Cloud for sharing
• 165 Million in US — employer covers health cost
• ACCESS to 2,000 clinical trials for repurposing

Constance Lehman Massachusetts General Hospital

• Pathway – Screening mammograms is run via the algorithm, report generated, Radiologist accept or reject the mobile presented
• more domains need AI and specialties of Medicine
• Development and implementation then dissemination to multiple sites
• In 1000 mammogram only 8 cancers detected

David Schenkein GOOGLE Venture (GV)

• Regeneron, Geisinger Health System, Amgen – Big data repositories are built
• Clinical Trials and big data
• delivery side: AI: Pathology and Radiology, early diagnostics:
• AI in MDs Offices to record the entire sessions
• diagnostics – Foundation Medicine diagnostics had difficult time to prove improvement to justify reimbursement

 

3:00–3:15 Break

 

3:15–4:40 Machine Learning into the Clinic

 

Moderator: Sangeeta Bhatia, Koch Institute, MIT

Tommi S. Jaakkola MIT Computer Science and Artificial Intelligence Laboratory

Machine learning for drug design

• Automation arises from data
• How much data is enough? assay size vs all spectrum
• MIT Consortium on Drug Discovery: 13 Pharma, chemistry Department, Chemical Engineering Department, CSMIT Consortium on Pharmaceutical Discovery
• multi-resolution representations
• substructures with precision attachment – what atoms are underlined
• predicting molecular properties
• programmatic optimization – machine translation problem in analogy – learn the mapping principles from examples
• Example comparisons – Programmatic optimization – Success %
• Understanding: Accuracy- interpretable, statistical constraints inserted
• integrations with biology and cancer research
• Automating Drug Design – learn from exampl

Andrew Beck PathAI

Artificial intelligence for pathology: From discovery to AI-powered companion diagnostics

• Pathologic diagnosis informs all subsequent medical decisions
• Discordance among pathologists is common in interpretation of breast biopsy
• Discordance among pathologists is common in interpretation of skin biopsy – melanoma
• Discordance among pathologists is common in interpretation across all fields: NSCLC, bladder,
• AI – deep learning is driving major advances in computer vision
• Reduce errors training set vs validation set
• computer vision vs Human eye – machine learning is BETTER
• Labeled and counting data standardization requires work with the community of Pathologists
• Automatic interpretation of cells and tissues
• Immuno-oncology
• TisSue and cellular phenologies – Pathological phenotypes
• BMS – CD8 continuum – Phenotype : Identify stromal and parenchymal CD8 infalmmation
• AI project with Novartis

Brian Wolpin Dana-Farber Cancer Institute

Early detection of pancreatic cancer

• ML for Pancreatic Cancer Early detection
• Metabolic alterations: Muscle wasting, subcutaneous adipose
• Automated Peripheral Tissue Segmentation
• 30,000 scans: changes by race, age, : Image-Based Risk Modeling within Large Health Systems

longitudinal Cohorts: Risk Factors

• Diabetes related to pancreatic cancer
• risk for weight and Diabetes
• Medication: Start and Stopped: NSAIDs, Tylenol, H2 Blocker, PPI, anti- HTN,
• ML – early detection: altered systemic metabolism

Stephen Friend University of Oxford

From complex to complicated problems: What are the known unknowns, and unknown unknowns in using wearables to forecast symptom transitions

• Pattern recognition for prognosis NEJM 20 years took since Breast cancer started to benefit from pattern recognition
• Participants-Centered
• 10M smart watches
• APPLE – many studies Smart watch – 15,000 participants donate data via their watch – large scale ALERT based on data analysis
• Inter-individual Diversity
• Personal Health Assistant
• all day sensing & recording
• Health Assessment – signals to Symptoms
• Deep Learning – multiple stages of non-linear feature transformation
• Vector Institute – data analysis was done [Pregnancy]
• Forecasting – binary labels: Well or Sick, intra and inter-individual longitudinal
• 4YouandMe  – Foundation
• Wearables for Cancer patients

4:40–4:45 Closing Remarks Phillip Sharp Koch Institute, MIT

 

 

Machine Learning and Cancer The 18th Annual Koch Institute Summer Symposium on June 14, 2019 at MIT’s Kresge Auditorium will focus on Machine Learning and Cancer. Both fields are undergoing dramatic changes, and their integration holds great promise for cancer research, diagnostics, and therapeutics. Cancer treatment and research have advanced rapidly with an increasing reliance on data-driven decisions. The volume, complexity, and diversity of research and clinical data—from genomics and single-cell molecular and image-based profiles to histopathology, clinical imaging, and medical records—far surpasses the capacity of individual scientists and physicians. However, they offer a remarkable opportunity to new approaches for data science and machine learning to provide holistic and intelligible interpretations to trained experts and patients alike. These advances will make it possible to provide far better diagnostics, discover possible chemical pathways for de novo synthesis of therapeutic compounds, predict accurately the risk of individuals for development of specific cancers years before metastatic spread, and determine the combination of agents that will stimulate immune rejection of a tumor or selectively induce the death of all cells in a tumor. The symposium will address these issues through three sessions: Machine Learning in Cancer Research: the Need and the Opportunity Machine Learning to Decipher Cellular and Molecular Mechanisms in Cancer Machine Learning into the Clinic Sessions will be followed by a panel discussion of broadly informed experts moderated by MIT President Emerita Susan Hockfield. Introductory remarks will be given by symposium co-chairs and Koch Institute faculty members Regina Barzilay, Aviv Regev and Phillip Sharp.   Keynote Speakers | Machine Learning in Cancer Research: the Need and the Opportunity James P. Allison, PhD MD Anderson Cancer Center Regina Barzilay, PhD MIT Computer Science and Artificial Intelligence Lab, Koch Institute for Integrative Cancer Research at MIT Aviv Regev, PhD Broad Institute, Koch Institute for Integrative Cancer Research at MIT   Session Speakers Michael R. Angelo, MD, PhD Stanford Unviersity Andrew Beck PathAI Stephen H. Friend, MD, PhD Sage Bionetworks Tommi Jaakkola, PhD MIT Computer Science and Artificial Intelligence Lab Dana Pe’er, PhD Memorial Sloan Kettering Cancer Center Peter Sorger, PhD Harvard Medical School Olga Troyanskaya, PhD Princeton University Brian Wolpin, MD Dana-Farber Cancer Institute   Panel Discussion | Big Data, Computation and the Future of Health Care James (Jay) Bradner, MD Novartis Clifford A. Hudis, MD American Society of Clinical Oncology Constance D. Lehman, MD, PhD Massachusetts General Hospital Norman (Ned) Sharpless, MD National Cancer Institute   Moderator: Susan Hockfield, PhD Koch Institute for Integrative Cancer Research at MIT [if gte mso 9]>

SOURCE

From: 2019 Koch Institute Symposium <ki-events@mit.edu>

Reply-To: <ki-events@mit.edu>

Date: Tuesday, March 12, 2019 at 11:30 AM

To: Aviva Lev-Ari <AvivaLev-Ari@alum.berkeley.edu>

Subject: Invitation to the 2019 Koch Institute Symposium – Machine Learning and Cancer