Can Automated AI Drug Discovery Platforms Increase Pharmaceutical Companies’ Confidence in AI Techniques (Receptor.AI Report)


LONDON, UK, Aug. 14, 2022 / — There is no doubt that artificial intelligence is a disruptive technology for drug discovery. The AlphaFold alone, which has de facto solved an eternal problem of protein structure prediction, is enough to affirm this. However, there is some deception in AI drug discovery techniques. Early adopters of AI technologies quickly realized that AI is not a silver bullet that will magically eliminate all the complexity and cost of the drug discovery process.

Receptor.AI has studied and spoken to many parties over the past few years who have used AI drug discovery tools and taken their feedback to help overcome these challenges with our solutions.

Early adopters found AI drug discovery workflow weak

The biggest hurdles to AI-based drug discovery are, predictably, data quality and poorly determined metrics. Even things seemingly as obvious as virtual screening become surprisingly complex if framed in terms of machine learning. We want to classify compounds into “effective” and “ineffective” for particular diseases, which looks like a trivial problem in ML, similar to finding images of cats among other images in computer vision.

However, unlike the images, which may or may not contain the cat’s head, we cannot even formalize what is “effective” when we talk about molecules. The ultimate goal is to find a compound that cures the disease, but getting enough training data for this parameter is physically difficult. This is why R&D teams are forced to use indirect endpoints, such as the strength of binding to a particular target protein.

In a way, all modern AI-based drug discovery is looking for ligands (something that binds) but not drugs (something that heals). Binding free energy (or binding score, which is an approximation of binding free energy) is a perfect measure for finding ligands, but not necessarily for finding effective drugs. Indeed, the perfectly binding ligand could fail miserably as a drug due to its toxicity, poor ADME properties, harmful off-target interactions, and a dozen other non-trivial reasons. That’s why Receptor.AI applies AI-based virtual screening and uses more than 40 predictive models, which estimate the multitude of biological factors, ranging from basic ADME-Tox/PK properties to assessment of off-target interactions. at the proteome level.

All of these filters are themselves AI models, which require reliable data. Lots of high quality data! Unfortunately, not all of the data accumulated by big pharma is directly usable to train AI models. Most of this data is generated by simple high-throughput tests capable of performing gargantuan amounts of parallel measurements. However, these measurements generally have poor predictive power in terms of the endpoint in vivo. AI models trained on this data show poor results not because the model architecture is bad (it is usually state-of-the-art) or the training data set is too small (it is actually huge), but simply because bad proxy endpoints are used.

Automation and personalization are key

It is also becoming increasingly clear that there is no such thing as a “one-size-fits-all drug discovery workflow”, commonly referred to as a “drug discovery pipeline” and visualized as a linear sequence of steps.
Based on discussions with several biotech and pharmaceutical companies, Receptor.AI arrived at the need for automation and precise control of every step of the AI ​​drug discovery workflow:
• Data preparation and quality assessment of training and test datasets.
• Correlation of proxy values ​​with biological parameters and selection of the most significant proxy representation.
• Intelligent AI assistant for data and model quality control.
• Mapping raw data to most significant proxy values.
• Choose the most appropriate AI model architecture and training protocol.
• Execution of model training, evaluation and adjustment.
• Management of different model versions and parameter sets.
• Deployment of the model and monitoring of its performance in real projects.
• Customizable AI pipeline to flexibly tackle specific drug discovery projects.

Better tools to overcome frustration

To solve most of these problems, Receptor.AI is currently developing the next-generation AI platform for drug discovery, which is not only powered by AI, but also automated and assisted by AI.

The platform is fully configurable and includes an easy-to-use pipeline builder, which allows the user to design their own drug discovery workflow. The platform automates and controls all routine drug discovery ML tasks focusing on successful drug design, not ligand design: data quality assessment, data filtering, property extraction and model architecture selection, continuous model retraining/tuning triggered by data or model architecture changes, model deployment, model performance testing, version control of model architectures and parameters, etc. .

Such a system solves the majority of technical problems encountered repeatedly by the AI ​​department of pharmaceutical companies in each new project, leading to a clean, efficient and organized working environment. Most routine tasks are not only automated but are going to be monitored by “advisory AI”, which adapts to the company’s internal workflow and balances tasks and resources accordingly.

Importantly, the platform could be deployed in two modes of operation:

1. Entirely on-premises, to ensure data protection, access rights and security for Big Pharma;

2. Cloud-based, which offers near-infinite scalability and accessibility for SMBs.
For now, all major elements of the platform are fully functional internally and are used in pilot projects, which Receptor.AI is carrying out in collaboration with several research institutes and CROs. The complete system can be deployed on-premises or in the secure cloud and tailored to the needs of the pharmaceutical company.

There is also a SaaS solution for ultra-fast virtual screening at multi-billion dollar scale (in less than an hour), which exposes the platform’s most popular modules through a user-friendly interface designed to be “a Google search for new hit compounds”. “. SaaS is most useful for small and medium-sized biotech companies and academic institutions, which need low-cost, hassle-free solutions for individual steps in the drug discovery workflow.

Alan Nafiyev
+ +49 15172837276
write to us here
Visit us on social media:


Comments are closed.