Recursion Pharmaceuticals, Inc. · NASDAQ:RXRX
Recursion Pharmaceuticals — Investor Day (June 2024)
179 slides from Recursion Pharmaceuticals's investor day presentation, Jun 24, 2024
Slide 1: Download Day 2024
Slide 2: Agenda placeholder
AGENDA PLACEHOLDER
Slide 3: Morning and afternoon sessions
Morning Session 9:30 am – 12:30 pm Afternoon Session 1:30 pm – 4:30 pm Welcome State of Recursion Chris Gibson PhD – Co-Founder & CEO Recursion OS Lina Nilsson PhD – Senior VP of Inception Labs Preclinical Laura Schaevitz PhD – Senior VP and Head of Research Fireside Chat with Deepak Nijhawan, MD, P
Slide 4: Welcome: State of the Company
= or Welcome State of the Company <BownloadDay 2024 =
Slide 5: Forward-looking statements
This presentation and any accompanying discussion and documents contain information that includes or is based upon "forward-looking statements" within the meaning of the Securities Litigation Reform Act of 1995. These forward-looking statements are based on our current expectations, estimates and pr
Slide 6: Goals for today
Help define what we view as a tipping point moment as BioTech transitions to TechBio and understand why Recursion is uniquely positioned to take advantage of this Share details and updates on our: • Pipeline – with 7 clinical trial readouts expected in the next ~18 months • Partnerships - with poten
Slide 7: The Moment: A Tale of Two Cities
The Moment: A Tale of Two Cities
Slide 8: Scannell et al. drug discovery productivity
Scannell et al., Nature Reviews Drug Discovery 11:191, 2012
Slide 9: A Tale of Two Cities: Tech
A Tale of Two Cities: Tech Computational capacity of the fastest supercomputers The number of floating-point operations® carried out per second by the fastest supercomputer in any given year. This is expressed in gigaFLOPS, equivalent to 10 floating-point operations per second. 1 billion 4100 milion
Slide 10: Data, compute, and automation shifting speed, cost, quality
Data, compute, and automation are shifting the speed, cost, and quality of novel insights today, and we are nearing the stage where we can harvest the earliest of this jump forward New types of companies have emerged that are truly “bilingual” in tech and science We believe the transformation of Bio
Slide 11: Most BioTech companies have built a point solution
Most BioTech companies have built a point solution - they’ve developed a tool, process, model or analysis to accomplish an important step in drug discovery. This is how we started too. But discovering and developing medicines requires hundreds of steps… Phenomics
Slide 12: To industrialize drug discovery, point solutions must be integrated
To truly industrialize drug discovery, point solutions must be integrated as modules across many diverse steps Patient Connectivity & Novelty Hit & Target Validation Compound Optimization Translation | SNDEnabling Studies] Clinical Development Invvomie Priontization ie v target predictions *Currenty
Slide 13: Recursion OS
~) LOWE
Slide 14: Data layers
Genomics Proteomics Phenomics ADME InVivomics ——H—_- Real-World Patient Data ~ DownloadDay 2024 @ Recursion
Slide 15: Connecting data layers to build digital maps of biology
Connecting data layers to build digital maps of biology Genoni¢es) A$AHaASHAH 0) Transcriptomics ————— Proteomie¢ss AtAMHmH (cea) Real-World Patient Data fo} Recursion
Slide 16: Data layers
Genomics Transcriptomics Proteomics ® © Phenomics ADME InVivomics Real-World Patient Data ~ DownloadDay 2024 @ Recursion
Slide 17: Connecting data layers to build digital maps of biology
Connecting data layers to build digital maps of biology Genomics Transcriptomics Proteomics 9 InVivomics (cee) Real-World Patient Data ~DownloadDay 2024 @ Recursion
Slide 18: Where are we right now?
Where are we right now? DownloadDay 2024 @ Recursion
Slide 19: More data and compute enable generalizable models
More data and compute enables more generalizable Models Fit for Purpose Data is Critical and the primary bottleneck More Compute is needed and few are investing at our scale
Slide 20: Pre- vs post-industrialization
Pre-Industrialization Post-Industrialization Team of 40 people 30 program hypotheses explored Team of 7 people 115 program hypotheses explored ~1 FTE equivalent 201 program hypotheses initiated 2022 2023 Half of 2024 Continued Improvement
Slide 21: Four ingredients needed to continue leading TechBio
Four ingredients needed to continue leading TechBio at the tipping point DownloadDay 2024 a Compute Capital @ Recursion
Slide 22: Four ingredients: People
Four ingredients needed to continue leading TechBio in an industry at the tipping point People
Slide 23: Four ingredients: Data
Four ingredients needed to continue leading TechBio in an industry at the tipping point Data
Slide 24: Four ingredients: Compute
Four ingredients needed to continue leading TechBio in an industry at the tipping point Compute
Slide 25: Four ingredients: Capital
Four ingredients needed to continue leading TechBio in an industry at the tipping point Capital
Slide 26: We exist to run an experiment
We exist to run an experiment... ….An experiment to determine if there might be a better way to discover and develop drugs... ...We need this sort of ambition in BioTech if we hope to have a chance of transforming our ability to impact patients and drive down the cost of medicines.
Slide 27: Platform, pipeline and partnership updates
Platform • Internal programs now initiated by LLM with multiple hit nominations for LLM-generated programs with more on the way • Moving towards large-scale multi-omics and generalizable foundation models with first genome-scale transcriptomics map and patient data • Data and tools available to biop
Slide 28: The Recursion Operating System
The Recursion Operating System
Slide 29: 29
29
Slide 30: Virtuous cycles for end-to-end drug discovery
Virtuous Cycles Connect Systems for Efficient End-to-End Drug Discovery Patient Connectivity & Novelty Hit & Target Validation Compound Optimization Translation [ SNDEnebiingStudIB=N Clinical Development Liw-based automated IND ‘goneration® ‘target predictions ‘curently being explored ~ DownloadDay
Slide 31: Virtuous cycles explore biological variability
Virtuous Cycles Enable Efficient Exploration of Biological Variability
Slide 32: Virtuous cycles explore biological variability
Virtuous Cycles Enable Efficient Exploration of Biological Variability
Slide 33: Virtuous cycles explore biological variability
Virtuous Cycles Enable Efficient Exploration of Biological Variability
Slide 34: Knowledge, AI, experiments cycle
Kn ow le dg e AI Strategy Experiments Randomized Experiments Experiments
Slide 35: Virtuous cycles drive superlinear knowledge creation
Virtuous Cycles Drive Superlinear Knowledge Creation DownloadDay 2024 700M 600M 500M 400 M Millions of Experiments 8 = 200M 100M 2021 2022 2023 2024 Future 77 6T sT at 3T aT 7 Trillions of Relationships fo} Recursion
Slide 36: Data generation, compute, models
Slide 37: Compute, data generation, models
9g Recursion. er Compute Compute Data Generation Models “BownloadDay 2024
Slide 38: BioHive-1 to BioHive-2 supercomputer
BioHive-1: • 320 NVIDIA A100's • #84 on TOP500 list when built Nvidia invests $50 million in biotech company Recursion for A.I. drug discovery BioHive-2: • 504 NVIDIA H100's, 4X BioHive-1 • Top ranked supercomputer in pharma • #35 on TOP500 list 2020 Initial discussions began 2021 2023 2024
Slide 39: Compute
DownloadDay 2024 i ay 202 @ Recursion
Slide 40: Biological recall vs model size
0.48 0.49 0.47 0.46 0.45 0.44 0.43 0.42 10⁸ 10⁹ Parameters Bi ol og ic al re ca ll DL-2 Phenom-Beta Phenom-1 0.50 Recall of biological relationships vs model size
Slide 41: 35th fastest supercomputer in the world
Slide 42: Data generation
9g Recursion. ~~ DownloadDay 2024
Slide 43: Standardizing and automating experiments
Standardizing and automating experiments to capture multiple layers of biology and chemistry Genomics ——__+ QO) Transcriptomics ———————————+ Proteomi¢es tt Phenomics ADME @ InVivomics. ————+ <—.$—— ~ DownloadDay 2024 @ Recursion Y
Slide 44: Phenomics data generation and models
Brightfield to capture dynamics DATA GENERATION MODELS increase in expressed gene knock-outs detected >250 million experiments >50 human cell types >1 trillion neurons generated 2 w eeks Parameters Bi ol og ic al re ca ll Recall of biological relationships vs model size 0.48 0.50 0.46 0.44 0.42 10⁸
Slide 45: Transcriptomics map and models
>1M samples sequenced First genome-scale transcriptomic map DATA GENERATION MODELS Ability to predict compounds that failed later disease-relevant assays in internal tests Ability to predict compounds that passed later disease-relevant assays in internal tests Replaced time-consuming, disease-specif
Slide 46: ADMET model performance and throughput
1.25 1.00 0.75 0.50 0.25 0.00 -0.25 10⁶ 10⁷ 10⁸ 10⁹ Generalized performance over all 22 TDC ADMET datasets Number of parameters N or m al ize d pe rf or m an ce DATA GENERATION MODELS Estimated 90x throughput over manual approach >750 compounds per week Our single generalizable model improves with m
Slide 47: Digital in vivo cages
DATA GENERATION MODELS • Machine learning enables scale by extracting signals from video and temperature sensors • Applied across breadth of Recursion portfolio • Designed to select the right molecule at the right dose before entering studies >1,000 digital mouse cages 150 digital rat cages in 2024
Slide 48: Real-world patient data
DATA GENERATION MODELS >20 PB of real-world multi-modal oncology data Hundreds of thousands of unique de-identified patient records across diverse therapeutic areas Combining Recursion maps of biology with patient clinical data unlocks causal modeling to find novel targets
Slide 49: Data layers
0) Transcriptomics ———— Proteomics Phenomics ADME InVivomics — Real-World Patient Data ~ DownloadDay 2024 @ Recursion
Slide 50: Connecting data layers to build digital maps of biology
Connecting data layers to build digital maps of biology Genonies) $A 0) Transcriptomics ————— Proteomi¢ss tAMHJH (cea) Real-World Patient Data fo} Recursion
Slide 51: Data layers
Genomics Proteomics Phenomics ADME InVivomics Real-World Patient Data ~ DownloadDay 2024 @ Recursion
Slide 52: Connecting data layers to build digital maps of biology
Connecting data layers to build digital maps of biology Genomics O) Transcriptomics Proteomics @) ADME @ InVivomics (cee) Real-World Patient Data ~DownloadDay 2024 ® @ Recursion
Slide 53: Products of the OS
of the OS
Slide 54: Recursion OS: multiple potential product verticals
The Recursion OS: Utility across multiple potential product verticals Commercial Models Interfaces & Tools Pipeline DownloadDay 2024 @ Recursion Compute Data Generation NN, Models
Slide 55: Phenom-Beta outperforms CellProfiler
Phenom-Beta, available on NVIDIA BioNeMo, outperforms open-source "gold standard" CellProfiler CellProfiler Phenom-Beta Phenom-1 Phenom-2 candidate Re la tiv e Re ca ll Im pr ov em en t 1.00 1.05 1.10 1.15
Slide 56: Interfaces and tools across drug discovery
Interfaces and Tools: bringing together modules spanning the drug discovery process oS Phenomics Compound Patient-specific Phenom-1 Chemical InVivomic selection with ML hypotheses tractability prioritization IN ™ © ® € LLMs & literature InVivomic Chemical property Predicted target Tempus data Phenom
Slide 57: LOWE
~) LOWE
Slide 58: Turning drug discovery into a search problem
Turning drug discovery into a search problem
Slide 59: Turning drug discovery into a search problem
Turning drug discovery into a search problem
Slide 60: Pipeline: industrialized workflows
Pipeline: connecting systems into Industrialized Workflows Patient Connectivity & Novelty Hit & Target Validation Compound Optimization Translation [ SNDEnebiingStudIB=N Clinical Development Invivomic Proetization Luneoasea automated ND generators MZ Digital Tlerabity ‘experiments LLW-based ‘automat
Slide 61: Preclinical: The Power of Prediction
Preclinical: The Power of Prediction
Slide 62: Traditional novel opportunity timeline
Novel opportunity 1-2 years Millions of $ Disease relevance ~One potential program
Slide 63: Weeks vs. years, thousands vs. millions
Novel opportunity + Disease relevance + Predicted chemical properties Weeks vs. years Thousands of $ vs. millions Hundreds of potential programs evaluated, we choose the best
Slide 64: Rapidly narrowing the funnel
Rapidly narrowing the funnel through a standardized approach
Slide 65: Accelerate development of drug candidates
Accelerate the development of high potential drug candidates
Slide 66: Industry vs. Recursion time and cost
1-2 years Millions of $ Industry Recursion Weeks vs. years Thousands of $ vs. millions
Slide 67: Identifying novel targets and optimizing novel chemistry
Identifying novel targets and optimizing novel chemistry
Slide 68: PBMC-derived fibrocyte assay
PBMC-derived fibrocyte assay • Human-PBMCs are differentiated to fibrocytes • Treatment with a control peptide gives desired impact on fibrocytes (control state) Promising hits from phenotypic screen Control State: 10 µg/mL control peptide Disease State: Undisclosed treatment Concentration (µM)
Slide 69: Phenotypic assay concentration response
CONFIDENTIAL INFORMATION OF RECURSION REC-A 50% Rescue 1.5 µM Representative concentration response curves (CRC) from the phenotypic assay Healthy Control State Disease State Core N H R1 R4 R5 Core N H R2 H R5 REC-B 50% Rescue ~10 µM Significant reduction of disease modifying activity
Slide 70: Target Epsilon biochemical confirmation
Target Epsilon KO Target Epsilon IC50 = 31.5 nM Target Epsilon IC50 = 3600 nM Target Epsilon IC50 = 106 nMCore N H R1 R4 R5 Core N H R3 R4 R5 Epsilon biochemical confirmation 2.5 µM 2.5 µM 2.5 µM n/a Core N H R7 R5 R2 REC-C REC-A REC-D REC-D REC-A REC-C Target Epsilon KO GeneCompounds
Slide 71: Phenomap correlates with tubulin inhibition
Compounds with <10 µM tubulin IC50 Tubulin Related Genes Compounds with >30 µM tubulin IC50 Tubulin Related Genes REC-C Core N H R3 R4 R5 REC-D Core N H R7 R5 R2 Phenomap correlates with tubulin polymerization inhibition assay
Slide 72: Optimizing novel chemical matter
Industrialized Drug Discovery: Optimizing novel chemical matter Patient Connectivity & Novelty Hit & Target Validation Compound Optimization ‘Translation [ iNDEnabling studies] Clinical Development Invivomie Priortization tenable reverse-genetic causal ML models redicted target bind => sod EF nema p
Slide 73: In silico novel target prediction
Accelerating to IND enabling studies through in silico novel target prediction
Slide 74: CDK12, RBM39, CDK13 relationships
CDK12RBM39CDK13 CDK12 RBM39 CDK13 Similar Opposite
Slide 75: REC-E concentration effects
CDK12RBM392.5 μM1.0 μM0.25 μM0.1 μMCDK13 Similar Opposite CDK12 RBM39 2.5 μM 1.0 μM 0.25 μM 0.1 μM CDK13 REC-E REC-E
Slide 76: CDK binding predicted by MatchMaker
0 2 4 6 REC-E REC-F SR-4835 THZ531 CDK8 CDK2 CDK20 CDK9 CDK12 CDK13 CDK4 CDK6 CDK7 In si lic o – bi nd in g z- sc or e Probability of CDK Binding as predicted by MatchMaker {Positive control} {Positive control}
Slide 77: CDK12 activity of hit compounds
REC-E REC-F SR-4835 THZ531 100 75 50 25 0 0.001 0.01 0.1 1 10 % CD K1 2 Ac tiv ity (n or m . t o ve hi cl e) CDK12 inhibitors Hit Compounds CDK12 Activity Concentration (µM) {Positive control} {Positive control}
Slide 78: Rapid identification of development candidate
Predictions and minimal standard experiments enabled rapid and efficient identification of development candidate Number of Molecules Profiled Initial Synthesis (or Design) ~1,000,000 to Advanced Candidate (9 months) in Sui66 | Synthesized compounds , er in silico assessment Phenomics: Profiled 955 D
Slide 79: Rapid in silico novel target identification
Rapid in silico novel target identification Patient Connectivity & Novelty Hit Target Validation Compound Optimization rrr Clinical Development fo” Compound selection nox ‘ath cexperments cine =f “© ‘a — © Va Chemical Wactainty we 2M physical compounds & ~366.compounds with target predictions *curre
Slide 80: Time from target to lead
Time from tar @ Recursion Industry
Slide 81: Connecting data layers end-to-end
Connecting data layers end-to-end improves quality and speed of insights
Slide 82: REC-G and REC-H target data
Target 1 Target 1 REC-G and REC-H at 1, 3, 10 µM REC-G and REC-H at 1, 3, 10 µM
Slide 83: Phenomics and transcriptomics confirmation screens
Phenomics Confirmation Screen Transcriptomics Confirmation Screen Disease Score Disease O ff Ta rg et O ff Ta rg et Disea Score Target 1 KO Healthy REC-G 1 µM REC-G 2.5 µM REC-G 10 µM
Slide 84: REC-G tumor growth inhibition
Vehicle REC-G 30 mg/kg REC-G 100 mg/kg REC-G 300 mg/kg 100 mg/kg b.i.d REC-G TGI: 66% 50 mg/kg b.i.d REC-G TGI: 16% Tumor Volume m m 3 Days Vehicle Sunitinib:pos control TGI: 105% Daily Motion
Slide 85: End-to-end automation delivers lead in 10 months
End-to-end automation drives significant efficiency gains to deliver lead in 10 months Patient Connectivity & Novelty Hit & Target Validation Compound Optimization Translation. [ INDEnsbiing Studies] clinical Development @@ —° we 2M physical ‘compounds & 368 compounds with ideatior target prediction
Slide 86: Identifying molecules for undruggable target
Identifying novel molecules for a previously undruggable target
Slide 87: REC-J concentration effects
Target KO REC-J 0.1 µM REC-J 0.25 µM REC-J 1.0 µM REC-J 2.5 µM REC-J 2.5 µM REC-J 1.0 µM REC-J 0.25 µM REC-J 0.1 µM Target KO
Slide 88: REC-J vs REC-K stabilized protein
REC-J EC50 = 100 µM REC-K EC50 = >10 µM 3 4 2 1 0 0.01 0.1 1 10 100 Concentration (µM) Am ou nt o f S ta bi liz ed P ro te in REC-J REC-K
Slide 89: Overcoming undruggable targets
Overcoming the hurdles of drug discovery: undruggable targets Patient Connectivity & Novelty Hit & Target Validation Compound Optimization Translation Cl ae ML modes rained ‘on Tempus data human cel types ‘enable reverse-genetic ‘causal ML models Automated generation E oe a= ==-0 | ‘chemical proper
Slide 90: Unexplored oncology
‘Unexplored oncology Our unique approach creates Pate novel opportunities ° Low oncology High oncology relevance relevance RXRX Data RXRX Data ° cra ae Explored oncology DownloadDay 2024 Public Data @ Recursion
Slide 91: High potential, thoroughly explored
High potential, thoroughly explored
Slide 92: Potential vs exploration quadrants
High potential, unexplored Low potential, thoroughly explored Low potential, unexplored
Slide 93: Unexplored oncology
Unexplored oncology Public Data Low oncology High oncology relevance relevance RXRX Data RXRX Data % 0 cyias ae Explored oncology Public Ds * ~ DownloadDay 2024 tlie Data @ Recursion
Slide 94: Turning this into a search problem
We are turning this into a search problem, evaluating new programs in bulk ~ DownloadDay 2024 @ Recursion
Slide 95: Industry vs. Recursion time and cost
1-2 years Millions of $ Industry Recursion Weeks vs. years Thousands of $ vs. millions
Slide 96: Fireside Chat: Dr Deepak Nijhawan
Fireside Chat: Dr Deepak Nijhawan Associate Professor in the Departments of Internal Medicine and Biochemistry at UT Southwestern Medical Center
Slide 97: Convocation
Convocation BownloadDay 2024 =
Slide 98: AI impact in healthcare
Patient AI impact in healthcare Patients are waiting
Slide 99: AI across the healthcare value chain
Prediction Diagnosis Therapeutic Generation Therapeutic Development Personalization & Access Organizational Productivity Achieve earlier diagnosis through AI-enabled diagnostic methods, biomarkers, clinical decision support and more Generate and optimize new molecules, biologics, cell therapies, etc
Slide 100: From traditional to AI-powered approaches
To From Industrialized, AI- powered discovery Multiomics-based, image-based, holistic ML, generative AI and physics- based design, optimization, predictive models Precision medicine, adaptive trial design RWE and ML-driven enrollment, improved likelihood of success, and faster Personalized medicine,
Slide 101: Outcomes focused approach
Time Pr og re ss Relentlessly outcomes focused vs. point solutions Investing to innovate where it matters the most Agility to adopt new waves of tech + bio innovation Bilingual talent and culture (science + AI)
Slide 102: AI-enabled value chain and foundation
AI and multiomics to identify novel targets and comprehensive biological pathways Dry-wet lab combo - Industrialized workflows Cutting edge AI/Machine Learning Connected, high quality data at scale (proprietary + public) World class compute Complementary partnerships Bilingual talent (AI + science)
Slide 103: Therapeutic hypothesis to digital commercialization
Dry-wet lab - Industrialized workflows Cutting edge AI/ML Connected, high quality data World class compute Complementary partnerships Bilingual talent (AI + science) Pi pe lin e (in te rn al & e xt er na l) Fo un da tio n AI & multiomic Digital In Vivo Gen 1.0: Repurposed molecules Gen 2.0: New drug
Slide 104: Pipeline and foundation
Pi pe lin e (in te rn al & e xt er na l) Fo un da tio n Novel biology: ü RBM39, 12+ in early and late discovery • Roche – GI Onc, Neuroscience ü Bayer – Oncology Repurposed molecules to novel molecules: ü RBM39, Epsilon, 12+ in early and late discovery, 60% using Gen AI ü Roche – RG1 ü 2.5x lower co
Slide 105: Novel biology and Gen 2.0 NCE
• Novel biology: RBM39 ML-enabled phenomics insights Gen 2.0 NCE: • RBM39 • ~2.5x lower cost to IND • ~3x faster time to lead Therapeutic hypothesis Molecular invention Precision dev. Commercial- ization Pi pe lin e (in te rn al & e xt er na l) Fo un da tio n Phenom-1 LOWE 50 PB+ BioHive-2 Cross-ind
Slide 106: Cost to IND and time to validated lead
Co st to IN D ($ M ) Ti m e to V al id at ed L ea d (m o) 0 5 10 15 20 25 30 Industry Recursion 0 5 10 15 20 25 30 35 Industry Recursion ~2.5x lower cost ~3x faster • Novel biology: RBM39 Gen 2.0 NCE: • RBM39 • ~2.5x lower cost to IND • ~3x faster time to lead Therapeutic hypothesis Molecular invent
Slide 107: Faster recruitment
• Novel biology: RBM39 Gen 2.0 NCE: • RBM39 • ~2.5x lower cost to IND • ~3x faster time to lead Therapeutic hypothesis Molecular invention Precision dev. Commercial- ization Pi pe lin e (in te rn al & e xt er na l) Fo un da tio n Phenom-1 50 PB+ BioHive-2 Cross-industry partnerships – Enamine, Tempu
Slide 108: Therapeutic hypothesis, molecular invention, precision development
Therapeutic hypothesis Molecular invention Precision development
Slide 109: Phenom-Beta on NVIDIA BioNeMo
Phenom-Beta Hosted on Nvidia BioNeMo platform 75% Masked 25% of all images 86M parameter model 50% of all images 307M parameter model Original Image Fo un da tio na l m od el s f or e xt er na l u se • Novel biology: RBM39 Gen 2.0 NCE: • RBM39 • ~2.5x lower cost to IND • ~3x faster time to lead Ther
Slide 110: AI across preclinical to commercialization
AI & multiomics- based novel targets and pathways Dry-wet lab combo - Industrialized workflows Cutting edge AI/ML Connected, high quality data World class compute Complementary partnerships Bilingual talent (AI + science) Pipeline: Primarily Gen AI based new drug-like molecules Digital twins optimiz
Slide 111: Strategic priorities
Enhance AI-Driven Chemistry Advance Preclinical and Clinical Stage Programs Innovate with AI across Clinical Development Continue Investment in Scalable Infrastructure – wet and dry lab Deliver on Strategic Partnerships Create additional SaaS opportunities to advance the creation of medicines
Slide 112: Partnerships
Partnerships
Slide 113: Pipeline, partnership and data strategy
Build internal pipeline in indications with potential for accelerated path to approval Pipeline Strategy Precision Oncology Rare Disease Partnership Strategy Partner in complex therapeutic areas requiring large financial commitment or competitive arbitrage Leverage partner knowledge and clinical dev
Slide 114: Partnership strategy
Partnership Strategy Partner in complex therapeutic areas requiring large financial commitment or competitive arbitrage Leverage partner knowledge and clinical development capabilities Undruggable Oncology Other large, intractable areas of biology (e.g., CV/Met) Neuroscience* Pipeline Partnerships D
Slide 115: Roche and Genentech collaborations
Trademarks are the property of their respective owners and used for informational purposes only. Undruggable oncology targets Announced Sept 2020 Amended Nov 2023 Neuroscience (and single oncology indication) Announced Dec 2021
Slide 116: Computational Sciences in Drug Development
Computational Sciences in Drug Development John Marioni, PhD FMedSci Senior Vice President & Head of Computational Sciences, gRED
Slide 117: AI in computational sciences: past, present, future
AI IN COMPUTATIONAL SCIENCES: PAST, PRESENT, AND FUTURE 117
Slide 118: AI in computational sciences: past, present, future
AI IN COMPUTATIONAL SCIENCES: PAST, PRESENT, AND FUTURE 118
Slide 119: Pijuan-Sala et al., Nature 2019
Pijuan-Sala, Griffiths, Guibentif, et al., Nature, 2019
Slide 120: Making sense of huge data volumes
We can generate huge amounts of data—from both healthy and perturbed conditions… but how will we make sense of these data and make predictions about perturbations we have not seen?
Slide 121: Cell types, genes, variant combinations
AI IN COMPUTATIONAL SCIENCES: PAST, PRESENT, AND FUTURE 121 CELL TYPES AND STATES 104 GENES 2x104 VARIANT COMBINATIONS GENE COMBINATIONS 101342.4x10 8
Slide 122: Foundation models and generative AI
This is one example where computational models, especially foundation models and generative AI can transform how we discover and develop medicines
Slide 123: gRED Computational Sciences vision
gRED Computational Sciences (gCS) seeks to make this vision a reality This is one example where computational models, especially foundation models and generative AI can transform how we discover and develop medicines
Slide 124: HOW?
HOW?
Slide 125: Lab in a loop
EXPERIMENT AI/ML DATA LAB LOOP IN A
Slide 126: AI in computational sciences
AI IN COMPUTATIONAL SCIENCES: PAST, PRESENT, AND FUTURE 126
Slide 127: Transformer-based oracle
EXPERIMENT AI/ML DATA TRANSFORMER BASED ORACLE
Slide 128: Our AI strategy for R&D
Our AI strategy for R&D Scale & resolutionLab in a Loop, integrated QUANTITY QUALITY Full stack, across all aspects of R&D; up to “self drive” Maximize benefit of large size: proprietary legacy data and data generation capacity Key partnerships: Strategic partnerships Partnership around unique data
Slide 129: Right target or chemical matter for the disease
RIGHT TARGET OR CHEMICAL MATTER FOR THE DISEASE
Slide 130: Experiment: perturbations and measurements
Experiment: perturbations Measurements RISPR Small GE") molecules < Imaging (—° Gastrointestinal cancer Neuroscience Biological Biological system 2 system 1 Lt xg Bo — Se || ete In collaboration with Target and compound prediction Query maps @ Recursion
Slide 131: Right molecule
RIGHT MOLECULE
Slide 132: Multi-modal model development
Cells Treatments scRNA-seq Multi-Modal Model Development Pheno-transcripto map Multi-modal embedding Cell painting Multi-modal data collection Cell painting embedding Cell painting Pheno-map scRNA-seq scRNA-seq embedding Transcripto-map Data generation 132 1 2 Validation + Discovery
Slide 133: Models are only as good as the data
MODELS ARE ONLY AS GOOD AS THE DATA
Slide 134: Challenges
Challenges Data management, metadata and access Integrating expertise from multiple disciplines Access to scalable scientific computing for fitting/fine-tuning models Democratizing access and ensuring use of data and models
Slide 135: Challenges and solutions
Challenges... But already driving to solutions Data management, metadata and access: modernizing our data stack and exploiting the cloud and associated tools Integrating expertise from multiple disciplines: internal organizational structure and external partners Access to scalable scientific computi
Slide 136: Thank you!
THANK YOU!
Slide 137: Roche and Genentech collaborations
Neuroscience (and single oncology indication) Announced Dec 2021 Undruggable oncology targets Collaboration announced Sept 2020 Amended Nov 2023 Trademarks are the property of their respective owners and used for informational purposes only.
Slide 138: Bayer collaboration timeline
Q1 Work initiated Initiated 1st joint Project which will now be advancing rapidly towards Lead Series nomination On track to complete 25 unique multi-modal data packages in Q3 Nov 2023 pivot to Oncology Q2 Q3 Bayer expected to be the first beta-user of our LOWE LLM- orchestrated workflow software
Slide 139: Scientific collaborations
Scientific Collaborations: Undruggable Or Decoding Biology ——— to Radically » y Lives DownloadDay 2024 GQ Recursion
Slide 140: Technology and data partners
Trademarks are the property of their respective owners and used for informational purposes only. Computation and ML/AI Real-world data access (oncology) Announced Nov 2023 Announced July 2023 Announced Dec 2023 Cheminformatics and chemical synthesis Announced May 2024 Real-world data access (non-onc
Slide 141: Real-world data network
• Multi-site network protocol continuously aggregating in various therapeutic areas • Geographically and demographically diverse population consented for re-contact • Whole exome sequencing paired with rich, longitudinal clinical data for all consenting patients • Access to hundreds of thousands of
Slide 142: Multi-pronged capital-efficient business strategy
We harness value from the Recursion OS with a multi-pronged capital- efficient business strategy ©. 20 (53) Pipeline Partnerships £ ~ DowntoadDay 2024 @ Recursion
Slide 143: Clinical programs
4 mn A i soe S&S =
Slide 144: Pipeline of programs
144 More than a dozen discovery and research programs in oncology or with our partners – first program optioned by Roche-Genentech in GI-oncology All populations defined above are US and EU5 incidence unless otherwise noted. EU5 is defined as France, Germany, Italy, Spain, and UK. (1) Prevalence for
Slide 145: REC-994 for CCM program summary
Target / MOA Superoxide Scavenger Molecule Type Small Molecule Lead Indication(s) Cerebral Cavernous Malformations Status Phase 2 Designation(s) US & EU Orphan Drug Source of Insight Recursion OS
Slide 146: CCM pathophysiology and key elements
PATHOPHYSIOLOGY & REASON TO BELIEVE PREVALENCE & STANDARD OF CARE CAUSE KEY ELEMENTS Symptomatic US + EU5, >1 million patients worldwide live with these lesions today LOF mutations in genes CCM1, CCM2 & CCM3, key for maintaining the structural integrity of the vasculature due to unknown mechanisms E
Slide 147: REC-994 preclinical studies
147 Source: Data above from Gibson, et al. Strategy for identifying repurposed drugs for the treatment of cerebral cavernous malformation. Circulation, 2015 or Recursion internal data (Ccm1 mouse model) Preclinical Studies: REC-994 reduces lesion burden and ameliorates vascular defects in genetic mo
Slide 148: SYCAMORE trial update
Source : The Symptomatic Cerebral Cavernous Malformation Trial of REC-994 (SYCAMORE) Screening & Randomization 1:1:1 Treatment Follow-up Outcome Measures Enrollment Criteria • MRI-confirmed CCM lesion(s) • Familial or sporadic • Symptoms directly related to CCM • Primary: Safety and tolerability • S
Slide 149: CCM trial outcome measures
Outcome Measures • Primary: Safety and Tolerability • Adverse events & symptoms • Secondary & Exploratory: • Efficacy § Clinician-measured outcomes (CGI, PGI) § MRI Imaging § Impact of acute stroke (mRS, NIHSS) § Patient and Investigator reported outcomes (SMSS, PROMIS-29, CCM-HI, symptom questionna
Slide 150: REC-2282 for NF2 meningiomas program summary
Target / MOA HDAC Inhibitor Molecule Type Small Molecule Lead Indication(s) NF2 Mutated Meningiomas Status Phase 2/3 Designation(s) Fast Track; US and EU Orphan Drug Source of Insight Recursion OS
Slide 151: NF2 pathophysiology and key elements
PATHOPHYSIOLOGY & REASON TO BELIEVE PREVALENCE & STANDARD OF CARE CAUSE KEY ELEMENTS LOF mutations in NF2 tumor suppressor gene, leading to deficiencies in the tumor suppressor protein merlin Efficacy signal in Recursion OS, cellular, and animal models; suppression of aberrant ERK, AKT, and S6 pathw
Slide 152: REC-2282 preclinical data
152 NF2 knockdown cells Healthy Cells REC-2282 HUVEC, human umbilical vein endothelial cells; NF2, neurofibromatosis type 2; siRNA, small interfering RNA. Prevents growth & regrowth of NF2- deficient meningioma model in mice + REC-2282 (months) + Normal diet (months) Clinical: NF2
Slide 153: POPLAR-NF2 trial design
Phase 2/3 trial initiated in Q2 2022 Outcome Measures Key Enrollment Criteria • MRI-confirmed progressive meningioma • Sporadic meningioma with confirmed NF2 mutation • Familial NF2 meningioma • Have documented progression with past 24 months • Primary: PFS6 defined as proportion of patients who are
Slide 154: REC-4881 for FAP program summary
Target / MOA MEK Inhibitor Molecule Type Small Molecule Lead Indication(s) Familial Adenomatous Polyposis Status Phase 2 Designation(s) Fast Track; US and EU Orphan Drug Source of Insight Recursion OS
Slide 155: FAP pathophysiology and key elements
Polyps Found in Colon and Upper GI Tract PATHOPHYSIOLOGY & REASON TO BELIEVE PREVALENCE & STANDARD OF CARE CAUSE KEY ELEMENTS Inactivating mutations in the tumor suppressor gene APC • Targeting classical FAP patients (with APC mutation) • MEK inhibitor, small molecule • Oral dosing • Preliminary rea
Slide 156: FAP preclinical validation
Clinical: FAP @ Preclinical Validation of Novel OS Insight in Relevant FAP Models REC-4881 rescued phenotypic defects of cells with APC knockdown wv polyp count 0.0001 8 S & = s & 3» $30: . = 8154 § 3 j § 20: a & @? oo 8 i & 10 Lo Cells 3 5 e 0 roan ves eee ee ee 4 os 3 os 7 1s 8 Disease Score Media
Slide 157: TUPELO trial design
157 Source : Evaluate REC-4881 in Patients With FAP (TUPELO) Part 2 Enrollment Commenced Outcome Measures Key Enrollment Criteria • Confirmed APC mutation • ≥ 55 years old • Post-colectomy/proctocolectomy • No cancer present • Polyps in either duodenum (including ampulla of vater) or rectum/pouch •
Slide 158: REC-4881 AXIN1/APC program summary
Target / MOA MEK Inhibitor Molecule Type Small Molecule Lead Indication(s) Solid Tumors with AXIN1 or APC Mutations Status Phase 2 Source of Insight Recursion OS
Slide 159: AXIN1/APC pathophysiology and key elements
PREVALENCE & STANDARD OF CARE CAUSE LOF mutations in AXIN1 or APC tumor suppressor genes PATHOPHYSIOLOGY & REASON TO BELIEVE Alterations in the WNT pathway are found in a wide variety of tumors and confer poor prognosis and resistance to standard of care Efficacy signal in the Recursion OS and favor
Slide 160: REC-4881 vs AXIN1 and APC knockouts
160 REC-4881 is phenotypically opposite to the genetic KO of APC and AXIN1 providing a novel mechanism that may restore the disease state modeled by the loss of these genes 0.1 µm 0.25 µm 1.0 µm 2.5 µm AXIN1 AXIN 1 APC APC 2.5 µm 1.0 µm 0.25 µm 0.1 µm REC 4881 Clinical: AXIN1 or APC Significantly gr
Slide 161: AXIN1/APC trial update
FPI achieved Q1 2024 Outcome Measures Enrollment Criteria • Unresectable, locally advanced, or metastatic cancers • ≥ 55 years old • AXIN1 or APC mutation confirmed by NGS (tissue or blood) • CRC patients must be RAS / RAF wildtype • No MEK inhibitor treatment within 2 months of initial dose • ≥ 1 p
Slide 162: REC-3964 for C. difficile program summary
Target / MOA Selective C. difficile Toxin Inhibitor Molecule Type Small Molecule Lead Indication(s) Prevention of CDI Status Phase 2 Source of Insight Recursion OS
Slide 163: C. difficile pathophysiology and treatment
163 PATHOPHYSIOLOGY & REASON TO BELIEVE Diagnosed US + EU5 patients • Severity of infection varies and can range from mild to severe, requiring colectomy • >29,000 patients die in the US each year from CDI • Cost burden of up to $4.8bn annually TREATMENT PARADIGM • Standard of care for 1st occurrenc
Slide 164: Clinical: C. difficile
Disease State Healthy Cells REC-3964 0.1 µM Clinical: C. difficile
Slide 165: C. difficile Phase 2 design
Screening Randomization & Treatment • Phase 1 and DDI studies completed • Phase 2 initiation expected in Q4 2024, preliminary readout expected by end of 2025 Trial Updates REC-3964 250 mg orally BID REC-3964 500 mg orally BID Observational R 2:1:1 N=80 High Risk of Recurrence Patients with confirmed
Slide 166: RBM39 degrader program summary
Target / MOA RBM39 Molecular Glue Degrader Molecule Type Small Molecule Lead Indication(s) TBD Status IND submission in Q3 2024, Phase 1/2 initiation in Q4 2024 Source of Insight Recursion OS
Slide 167: Planned Phase 1/2 RBM39 study
Planned Phase 1/2 study of RBM39 degrader in Biomarker Selected Relapsed Refractory HR-Proficient Solid Tumors Phase 1A Dose Finding (Accelerated Titration BOIN) N=30 Dose Level 1 Dose Level 2 Dose Level 3 Dose Level 4+ Single Patient, Accelerated Titrations until Pharmacologically Active Dose (PAD)
Slide 168: Integrating data and AI across clinical development
to truly industrialize drug discovery, data and AI solutions must be integrated as modules across many steps Exciting scientific collaborations span biopharma, tech & data Clinical Development
Slide 169: Clinical development steps
Treatments begin Clinical Development Good Lab Practices (GLP) & Toxicity Chemistry, Manufacturing, and Controls (CMC) Patient Selection/ Biomarker Strategy Trial Design/ Statistical Approach Trial Start-up Enrollment
Slide 170: Reducing time from IND-enabling studies to first patient
Treatments begin Clinical Development Good Lab Practices (GLP) & Toxicity Chemistry, Manufacturing, and Controls (CMC) Patient Selection/ Biomarker Strategy Trial Design/ Statistical Approach Trial Start-up Enrollment Collectively we believe these processes could significantly reduce time from IND e
Slide 171: Company & Milestones
Company & Milestones
Slide 172: Decoding Biology to Radically Improve Lives
Decoding Biology to Radically Improve Lives PRINCIPLES Explore the Uncharted Create Virtuous Cycles Build Connected Data Industrialize to Scale Act Boldly with Integrity Optim or rtfolio We are One Re n Challenge DownloadDay 2024 @ Recursion
Slide 173: Workforce and locations
~43% Female Male ~55% ~1% Non-Binary Parity Pledge Signer gender parity and people of color parity Data shown reflective of Q1 2024, gender statistics include participating individuals have advanced degrees >50% Technology – data science, software engineering, automation, etc. Life Sciences – biolog
Slide 174: Inception, Accelerator, OS and Value
Inception / Valence De-risks novel capabilities to expand the Recursion OS to the point of minimum viable product Accelerator Drives company-wide efficiencies and paves the way for achieving the Recursion mission OS Develops, deploys, and improves the platform and workflows that make up the Recursio
Slide 175: Pipeline milestones
Pipeline • CCM: Ph2 readout expected in September 2024 • NF2: Ph2 safety & preliminary efficacy expected in Q4 2024 • FAP: Ph2 safety & preliminary efficacy expected in H1 2025 • AXIN1 or APC Mutant Cancers: Ph2 FPI achieved in Q1 2024 with safety & preliminary efficacy expected in H1 2025 • C. diff
Slide 176: Partnership milestones
Partnerships • Roche & Genentech: validation program option exercised for 1st validated hit series in oncology, potential program & map options on the near or very near-term • Bayer: delivered multiple oncology data packages, on track to complete 25 unique data packages in Q3 2024, initiated and adv
Slide 177: Fireside Chat with Jensen Huang
Fireside Chat with Jensen Huang
Slide 178: Closing Remarks
Closing Remarks
Slide 179: Goals for today
Help define what we view as a tipping point moment as BioTech transitions to TechBio and understand why Recursion is uniquely positioned to take advantage of this Share details and updates on our: • Pipeline – with 7 clinical trial readouts expected in the next ~18 months • Partnerships - with poten