We send the initial Global Model weights to companies in the consortium.
Using our Global Model weights, our pharma partners train the remote local models.
Local models send gradients back to the server.
We aggregate those model weights.
We send aggregated model weights back to our pharma partners for local model training.
This process repeats back and forward as long as it takes until the model converges.
After that, our server has the converged global model.
We then send the converged model to our partner to deploy in their discovery efforts.
1
2
3
4
5
6
7
8
/ 8
As part of a pre-competitive industry consortium, federated learning enables fine-tuning of SOTA AI models on blinded
data sets from multiple Pharma companies.
Private data never leaves the local devices or Partner's data centers
Federated Learning is an ML technology that enables participants to build a high-performing global model without ever sharing their dataset.
A learning paradigm seeking to address the problem of data governance and privacy by training algorithms collaboratively without exchanging the data itself
Allows to build a model from decentralized data
Unlocks and increases the collaboration opportunities between partners even when they are competitors
We send the initial Global Model weights to companies in the consortium.
2
/ 8
Using our Global Model weights, our pharma partners train the remote local models.
3
/ 8
The local model gradients are sent back to our server
4
/ 8
We aggregate those model weights
5
/ 8
We send aggregated model weights back to our pharma partners for local model training.
6
/ 8
This process repeats until the global model converges.
7
/ 8
We then send the converged model to our partner to deploy in their discovery efforts.
Back to Platform Page
AI-driven Molecular Design & Optimization
AI-driven Molecular Design and Optimization
We leverage wet lab ground truth data – biopharma's gold standard in vitro, in vivo, and clinical experimental data.
Our robotics-driven automation facility enables us to design and scale ground truth assays for validating AI-designed molecules by integrating massive data, AI, and lab automation in a closed loop.
To overcome data sparsity barriers we augment it with two additional data streams.
We design ultra-high-throughput surrogate biological assays that can replace low-throughput gold standard assays. These “proxy” assays deconstruct biological processes into building blocks that are key for AI model training. Our invention replaces low throughput gold standard assays, which typically produce tens of data points, with proxy assays that can generate up to 100,000,000,000 data points in one week.
Computational Data
We use computational simulations including quantum mechanics, molecular mechanics, and thermodynamics to model biological processes, and generate billions of computational data points for AI model training. We also leverage generative AI to create synthetic data for model training.
Data Orchestration
With massive multimodal data, the Data Orchestration feature of our ITO™ platform formats, analyzes, cleans, annotates, curates, and featurizes our multimodal data prior to ingestion into our AI/ML models.
1. Format
5. Cleaning
2. Size
6. Annotation
3. Storage
7. Curation
4. Analytics
8. Featurization
Data Privacy, IP Protection, Security, Federated Learning
Model Foundry™
Our SOTA Model Foundry™ consists of 100’s of AI/ML models and a Multi-AI Agent System, with frontier models, each working to solve the multi-parameter optimization problem of drug discovery.
At the Transform step, our ITO™ platform also incorporates features for AI/ML models including evaluation, selection, sampling, training, inference, observability, molecule provenance, alignment, and safety.
Small Molecule Vs. Large Molecule
Once the right AI/ML model from our Model Foundry™ has ingested appropriate data for the product being designed (i.e. small molecule vs. large molecule), the output is an in silico molecule for which our ITO™ platform suggests the most probably synthetic routes to achieve feasibility, yield, and scale-up.
Output
The output stage begins with the receipt of a manufactured small or large molecule candidate. We leverage our SOTA robotics-driven lab facility to design and scale wet lab ground truth biochemical, cellular, and pharmacological in vitro assays for testing the small and large molecule therapeutic candidates.
To the Clinic
Manufacturing
Feasibility
Synthetic yield
Scale-up
CMC QC, QA
Small vs large
Tech Partnership
Our end-to-end ITO™ platform is a cloud-based system built on Microsoft Azure, leveraging our partnership with Microsoft Azure Quantum Elements. Through our partnerships with Microsoft and NVIDIA, our ITO™ platform deploys multi-GPU compute to ensure low latency and high throughput for our workflows.
HPC
AI
Future QC
Azure Quantum
Elements
IT infrastructure
Microsoft
Cloud Services
Trust
Security
Compliance
Supplies GPU and FLARE
Supplies GPU and FLARE
The platform can be replicated in other major cloud providers including Amazon Web Services (AWS), Google Cloud, Oracle Cloud and more...
AI-driven Molecular Design & Optimization
AI-driven Molecular Design & Optimization
We leverage wet lab ground truth data – biopharma's gold standard in vitro, in vivo, and clinical experimental data.
Our robotics-driven automation facility enables us to design and scale ground truth assays for validating AI-designed molecules by integrating massive data, AI, and lab automation in a closed loop.
To overcome data sparsity barriers, we augment it with two additional data streams.
We design ultra-high-throughput surrogate biological assays that can replace low-throughput gold standard assays. These “proxy” assays deconstruct biological processes into building blocks that are key for AI model training. Our invention replaces low throughput gold standard assays, which typically produce tens of data points, with proxy assays that can generate up to 100,000,000,000 data points in one week.
mRNA Display
Droplet Microfluidics
Phage Display
Single Cell Sequencing
Next Generation Sequencig (NGS)
Mass Spectrometry
Liquid Chromatography
Computational Data
We use computational simulations including quantum mechanics, molecular mechanics, and thermodynamics to model biological processes, and generate billions of computational data points for AI model training. We also leverage generative AI to create synthetic data for model training.
Quantum Mechanics
Molecular Dynamics
Thermo- dynamics
Synthetic Data
Data Orchestration
With massive multimodal data, the Data Orchestration feature of our ITO™ platform formats, analyzes, cleans, annotates, curates, and featurizes our multimodal data prior to ingestion into our AI/ML models.
1. Format
5. Cleaning
2. Size
6. Annotation
3. Storage
7. Curation
4. Analytics
8. Featurization
Private data never leaves the local devices or PharmaCos' data centers
2
Transform
SOTA Multi AI Agent System
Model Foundry™
Our SOTA Model Foundry™ consists of 100’s of AI/ML models and a Multi-AI Agent System, with frontier models, each working to solve the multi-parameter optimization problem of drug discovery.
At the Transform step, our ITO™ platform also incorporates features for AI/ML models including evaluation, selection, sampling, training, inference, observability, molecule provenance, alignment, and safety.
Small Molecule Vs. Large Molecule
Once the right AI/ML model from our Model Foundry™ has ingested appropriate data for the product being designed (i.e. small molecule vs. large molecule), the output is an in silico molecule for which our ITO™ platform suggests the most probably synthetic routes to achieve feasibility, yield, and scale-up.
Manufacturing
Feasibility
Synthetic yield
Scale-up
CMC QC, QA
Small vs large
3
Output
Output
The output stage begins with the receipt of a manufactured small or large molecule candidate. We leverage our SOTA robotics-driven lab facility to design and scale wet lab ground truth biochemical, cellular, and pharmacological in vitro assays for testing the small and large molecule therapeutic candidates.
To the Clinic
Tech Partnership
Our end-to-end ITO™ platform is a cloud-based system built on Microsoft Azure, leveraging our partnership with Microsoft Azure Quantum Elements. Through our partnerships with Microsoft and NVIDIA, our ITO™ platform deploys multi-GPU compute to ensure low latency and high throughput for our workflows.
IT infrastructure
Azure Quantum
Elements
HPC
AI
Future QC
Microsoft
Cloud Services
Trust
Security
Compliance
The platform can be replicated in other major cloud providers including Amazon Web Services (AWS), Google Cloud, Oracle Cloud and more...
Full ITO™ Platform
Watch how ITO™ works
Back to Platform Page
Federated Learning of AI-driven Precision Target ID and AI-driven Molecular Design & Optimization
Federated Learning
We offer AI infrastructure where biotech and pharmaceutical partners can build verticalized AI solutions across all modalities and therapeutic areas with federated learning.
Federated Learning of AI-driven Precision Target ID and AI-driven...
Federated Learning
We offer AI infrastructure where biotech and pharmaceutical partners can build verticalized AI solutions across all modalities and therapeutic areas with federated learning.
We leverage massive multiomic patient-derived tissue and clinical data to build frontier AI models that identify novel biological targets for molecular design.
Visualize in-patient and ambulatory data alongside a population-level genomic overlay to associate variants, genes, pathways, and proteins.
Data Cleansing & Biocuration
Panoramics
A domain-specific semantic layer designed primarily for biomedical and biological entities and their relationships.
2
Transform
In Silico Validation with Omics Evidence Display
Knowledge Graph Construction
Multimodal AI Models
Novel Gene Target Disease Associations
In Silico Validation with Omics Evidence Display
In Silico GDA Exploration with In Silico Discovery Query Tool
Graph Machine Learning for Hypothesis Generation
Clinical Expert Validation
3
Output
Novel Targets
To the Clinic
Back to Platform
AI-driven Precision Target ID
AI-driven Precision Target ID
We leverage massive multiomic patient-derived tissue and clinical data to build frontier AI models that identify novel biological targets for molecular design.
Visualize in-patient and ambulatory data alongside a population-level genomic overlay to associate variants, genes, pathways, and proteins.
Data Cleansing & Biocuration
Panoramics
A domain-specific semantic layer designed primarily for biomedical and biological entities and their relationships.
2
Transform
In Silico Validation with Omics Evidence Display
Knowledge Graph Construction
Multimodal AI Models
Novel Gene Target Disease Associations
In Silico Validation with Omics Evidence Display
In Silico GDA Exploration with In Silico Discovery Query Tool
Graph Machine Learning for Hypothesis Generation
Clinical Expert Validation
3
Output
Novel Targets
To the Clinic
Back to Platform Page
ITO™ Platform
We integrate massive multimodal data, frontier AI models, and high-throughput lab automation to identify novel disease targets and design small and large molecule therapeutics better and faster than traditional approaches.
Select the capability
AI-driven Precision Target ID
AI-driven Precision Target ID
AI-driven Molecular Design & Optimization
AI-driven Molecular Design & Optimization
Federated Learning
Federated Learning of AI-driven Precision Target ID and AI-driven Molecular Design & Optimization
