The Curing AI for Precision Medicine. Hoifung Poon

Transcription

1 The Curing AI for Precision Medicine Hoifung Poon 1

2 Medicine Today Is Imprecise Top 20 drugs 80% non-responders Wasted 1/3 health spending $750 billion / year 2

3 Disruption 1: Big Data : 40% 93% 3

4 Disruption 2: Pay-for-Performance Goal: 75% by

5 Vemurafenib on BRAF-V600 Melanoma Before Treatment 15 Weeks 5

6 Vemurafenib on BRAF-V600 Melanoma Before Treatment 15 Weeks 23 Weeks 6

7 Why We Haven t Solved Precision Medicine? ATTCGGATATTTAAGGC ATTCGGGTATTTAAGCC ATTCGGATATTTAAGGC ATTCGGGTATTTAAGCC ATTCGGATATTTAAGGC ATTCGGGTATTTAAGCC High-Throughput Data Discovery Bottleneck #1: Knowledge Bottleneck #2: Reasoning AI is the key to overcome these bottlenecks 7

8 Use Case: Molecular Tumor Board 8

9 Use Case: Molecular Tumor Board Problem: Hard to scale U.S. 2015: 1.6 million new cases, 600K deaths 902 cancer hospitals Memorial Sloan Kettering 2016: Sequence: Tens of thousand Board can review: A few hundred Wanted: Decision support for cancer precision medicine 9

10 First-Generation Molecular Tumor Board Knowledge bottleneck E.g., given a tumor sequence, determine: What genes and mutations are important What drugs might be applicable Can do manually but hard to scale 10

11 Next-Generation Molecular Tumor Board Reasoning bottleneck E.g., personalize drug combinations Can t do manually, ever 11

12 Big Medical Data Decision Support Precision Medicine Machine Reading Predict Drug Combo 12

13 13

14 PubMed 26 millions abstracts Two new abstracts every minute Adds over one million every year 14

15 Machine Reading PMID: 123 VDR+ binds to SMAD3 to form PMID: 456 JUN expression is induced by SMAD3/4 Knowledge Base 15

16 Machine Reading Involvement of p70(s6)-kinase activation in IL-10 up-regulation in human monocytes by gp41 envelope protein of human immunodeficiency virus type

17 Machine Reading Involvement of p70(s6)-kinase activation in IL-10 up-regulation in human monocytes by gp41 envelope protein of human immunodeficiency virus type 1... IL-10 GENE gp41 GENE human monocyte p70(s6)-kinase GENE 17 CELL

18 Machine Reading Involvement of p70(s6)-kinase activation in IL-10 up-regulation in human monocytes by gp41 envelope protein of human immunodeficiency virus type 1... Involvement REGULATION Theme Cause up-regulation REGULATION activation REGULATION Theme Cause Site Theme IL-10 GENE gp41 GENE human monocyte CELL p70(s6)-kinase GENE 18

19 Long Tail of Variations TP53 inhibits BCL2. Tumor suppressor P53 down-regulates the activity of BCL-2 proteins. BCL2 transcription is suppressed by P53 expression. The inhibition of B-cell CLL/Lymphoma 2 expression by TP53 negative regulation 532 inhibited, 252 inhibition, 218 inhibit, 207 blocked, 175 inhibits, 157 decreased, 156 reduced, 112 suppressed, 108 decrease, 86 inhibitor, 81 Inhibition, 68 inhibitors, 67 abolished, 66 suppress, 65 block, 63 prevented, 48 suppression, 47 blocks, 44 inhibiting, 42 loss, 39 impaired, 38 reduction, 32 down-regulated, 29 abrogated, 27 prevents, 27 attenuated, 26 repression, 26 decreases, 26 down-regulation, 25 diminished, 25 downregulated, 25 suppresses, 22 interfere, 21 absence, 21 repress 19

20 Machine Reading Prior work Focused on Newswire / Web Popular entities and facts Redundancy Simple methods often suffice High-value verticals Healthcare, finance, law, etc. Little redundancy: Rare entities and facts Novel challenges require sophisticated NLP 20

21 Precision Medicine Challenges Machine Reading Advances Annotation Bottleneck Distant Supervision Complex Knowledge Grounded Semantic Parsing Reasoning Beyond Sentence Boundary Neural Embedding Distant Supervision with Discourse Modeling 21

22 Free Lunch: Existing KB NCI Pathway KB Regulation Theme Cause Positive A2M FOXO1 Positive ABCB1 TP53 Negative BCL2 TP53 22

23 Free Lunch: Existing KB NCI Pathway KB Regulation Theme Cause Positive A2M FOXO1 Positive ABCB1 TP53 Negative BCL2 TP53 23

24 Free Lunch: Existing KB NCI Pathway KB Regulation Theme Cause Positive A2M FOXO1 Positive ABCB1 TP53 Negative BCL2 TP53 TP53 inhibits BCL2. Tumor suppressor P53 down-regulates the activity of BCL-2 proteins. BCL2 transcription is suppressed by P53 expression. The inhibition of B-cell CLL/Lymphoma 2 expression by TP53 24

25 Free Lunch: Existing KB NCI Pathway KB Regulation Theme Cause Positive A2M FOXO1 Positive ABCB1 TP53 Negative BCL2 TP53 TP53 inhibits BCL2. Tumor suppressor P53 down-regulates the activity of BCL-2 proteins. BCL2 transcription is suppressed by P53 expression. The inhibition of B-cell CLL/Lymphoma 2 expression by TP53 25

26 Free Lunch: Existing KB NCI Pathway KB Regulation Theme Cause Positive A2M FOXO1 Positive ABCB1 TP53 Negative BCL2 TP53 TP53 inhibits BCL2. Tumor suppressor P53 down-regulates Distant the activity Supervision of BCL-2 proteins. BCL2 transcription is suppressed by P53 expression. The inhibition of B-cell CLL/Lymphoma 2 expression by TP53 26

27 Genetic Pathways PubMed-scale extraction 15,000 genes, 1.5 million unique regulations Compare w. NCI: 10,000 unique regulations Applications UCSC Genome Browser, MSR Interactions Track U. Wisconsin breast cancer study Etc. Poon, Toutanova, Quirk, Distant Supervision for Cancer Pathway Extraction from Text. PSB

28 Complex Knowledge Outperform 19 out of 24 supervised participants in GENIA Shared Task Parikh, Poon, Toutanova. Grounded Semantic Parsing for Complex Knowledge Extraction, NAACL

29 Cross-Sentence, N-ary Relations The deletion mutation on exon-19 of EGFR gene was present in 16 patients, while the L858E point mutation on exon-21 was noted in 10. All patients were treated with gefitinib and showed a partial response. 29

30 Drug-Gene Interactions Distant supervision w. discourse modeling Orders of magnitude increase: ,952 No need for annotated examples Quirk & Poon, Distant Supervision for Relation Extraction beyond the Sentence Boundary, arxiv.org/abs/

31 Reasoning: Neural Embedding Embed gene network Entity / Relation (v 1, v 2,, v k ) Relation triple <subj, rel, obj> Score Distant supervision: Known relations score higher Increased recall by 20 points on held-out Toutanova et al., Representing Text for Joint Embedding of Text and Knowledge Bases, EMNLP-15. Toutanova et al., Compositional Learning of Embeddings for Relation Paths in Knowledge Bases and Text, ACL

32 Big Medical Data Decision Support Precision Medicine Predict Drug Combo 32

33 Drug Combination Problem: What combos to try? Cancer drug: 250+ approved, developing Pairwise: 719,400; three-way: 287,280,400 Wanted: Prioritize drug combos in silico 33

34 Drug Combination Problem: What combos to try? Cancer drug: 250+ approved, developing Pairwise: 719,400; three-way: 287,280,400 Wanted: Prioritize drug combos in silico Drug 1 Drug 2 34

35 35

36 BeatAML Targeted drugs: 149 Pairs: 11,026 Tested: 102; Unknown: 10,924 36

37 Machine Learning Evaluation: Cell kill + Synergy Learning: Ranking loss Features Panomics: Gene expression, Pharmacology: Drug targets Network knowledge: TP53 inhibits BCL2, 37

38 Interpretable Model Feature Weight BCL2 and MAPK3 (moderate) MAP2K1 or MAPK10 (moderate) BCL2 or MAPK3 (moderate) CSNK1E or PLK4 (high) MAP2K7 and MAPK7 (moderate) AKT3 and MAP2K1 (high) NEK2 or PLK1 (moderate) PSMB1 or PSMB2 (moderate) MAPK9 and STK11 (high) MAPK1 and MAPK13 (moderate) BIRC5 or PLK4 (moderate) MAP2K2 or MAPK14 (high) AKT3 and MAPK8 (moderate) STK10 and STK33 (high) BCL2 or MAPK8 (moderate) EGFR and MAPK3 (moderate) MAPK10 and MAPK3 (moderate) MAP2K1 and MAPK10 (moderate) BCL2 or MAPK1 (high) BCL2 or MAPK8 (high) Composite metric Feature Weight BCL2 or MAPK3 (moderate) AKT2 and BCL2 (high) MAP2K1 or MAPK10 (moderate) BCL2 and MAPK3 (moderate) MAP2K5 and MAPK4 (high) BCL2 and MAPK1 (high) MAPK4 and SRC (high) MAP2K2 or MAPK10 (moderate) AKT3 and MAP2K1 (high) BCL2 and MAPK3 (high) CSNK1D or PLK4 (moderate) MAP2K1 or MAPK13 (moderate) STK10 and STK33 (high) AKT3 and MAPK8 (moderate) MAPK3 and SRC (high) BIRC5 or PLK4 (moderate) MAP2K1 and MAPK10 (moderate) PLK1 and TAOK1 (moderate) MAP2K2 or MAPK14 (high) AURKB or PLK4 (moderate) AA metric 38

39 Interpretable Model Feature Weight BCL2 and MAPK3 (moderate) MAP2K1 or MAPK10 (moderate) BCL2 or MAPK3 (moderate) CSNK1E or PLK4 (high) MAP2K7 and MAPK7 (moderate) AKT3 and MAP2K1 (high) NEK2 or PLK1 (moderate) PSMB1 or PSMB2 (moderate) MAPK9 and STK11 (high) MAPK1 and MAPK13 (moderate) BIRC5 or PLK4 (moderate) MAP2K2 or MAPK14 (high) AKT3 and MAPK8 (moderate) STK10 and STK33 (high) BCL2 or MAPK8 (moderate) EGFR and MAPK3 (moderate) MAPK10 and MAPK3 (moderate) MAP2K1 and MAPK10 (moderate) BCL2 or MAPK1 (high) BCL2 or MAPK8 (high) Hanover: BCL2 + MEK Feature Weight BCL2 or MAPK3 (moderate) AKT2 and BCL2 (high) MAP2K1 or MAPK10 (moderate) BCL2 and MAPK3 (moderate) MAP2K5 and MAPK4 (high) BCL2 and MAPK1 (high) MAPK4 and SRC (high) MAP2K2 or MAPK10 (moderate) AKT3 and MAP2K1 (high) BCL2 and MAPK3 (high) CSNK1D or PLK4 (moderate) MAP2K1 or MAPK13 (moderate) STK10 and STK33 (high) AKT3 and MAPK8 (moderate) MAPK3 and SRC (high) BIRC5 or PLK4 (moderate) MAP2K1 and MAPK10 (moderate) PLK1 and TAOK1 (moderate) MAP2K2 or MAPK14 (high) AURKB or PLK4 (moderate) Impending trial on Venetoclax / Trametinib Composite metric AA metric 39

40 Project Hanover 40

41 Knowledge Machine Reading Can be done manually, need automation to scale E.g., PubMed search Reasoning Predictive Analytics Can t be done manually, need automation to enable E.g., personalize drug combinations 41

42 Collaborators Knight: Brian Drucker, Jeff Tyner Chicago: Andrey Rzhetsky Wisconsin: Mark Craven Microsoft Research: Chris Quirk, Kristina Toutanova, Scott Yih, Bill Bolosky, David Heckerman, Lucy Vanderwende, Ravi Pandya Interns: Maxim Grechkin, Ankur Parikh, Victoria Lin, Sheng Wang, Stephen Mayhew, Daniel Fried, Violet Peng, Hao Cheng 42