Planning and execution of chemical synthesis: tools and startups

Drug discovery is the process by which new drugs are discovered and involves: 🔹the identification of screening hits (high throughput screening in vitro and secondary assays), 🔹medicinal chemistry (design and synthesis) and 🔹optimisation of those hits to increase the affinity, selectivity, efficacy, metabolic stability and oral bioavailability. Plus in silico studies in combination with cellular functional tests are used to improve the functional properties of the drug candidates. Once a candidate compound that fulfills all of these requirements has been identified, the process of drug development can continue with preclinical and clinical trials.

In the the previous article I talked about some of the different AI tools and startups for primary and secondary screening during drug discovery. So, today I am going to talk about AI tools and startups for planning and execution of chemical synthesis during drug discovery.

🎯 AI in planning of chemical synthesis during drug discovery ⚗

Once a candidate molecule (screening hit) has been identified during-primary and secondary screening, then the drug discovery process proceeds with searching for an optimal chemical synthesis pathway to synthesise that drug candidate. Chemical synthesis is the artificial execution of useful chemical reactions to obtain one or several products.

Let’s see how AI can help here.

Retrosynthetic analysis (retrosynthesis) is a technique for planning the synthesis of a complex target molecule, by reducing the target into a sequence of progressively simpler structures (in branch points or on rings as they usually give straight chain fragments) that ultimately leads to the identification of commercially available starting material from which a chemical synthesis can then be developed.

Previously, algorithms like computer-assisted synthesis planning CASP,were used (that dates back to the 1960s) to assist retrosynthesis. This algorithm was focused on accelerating the process by which chemists decide how to synthesise small molecule compounds, but in the end failed to gain wide popularity among chemists.

So now, novel ML approaches are trained on empirical data and are used to predict the probability of the transformation of a particular branching position.

At each step, the target molecule (or an intermediate) can be linked to specific precursors via a predefined transformation rule, while AI algorithms can be trained from the literature regarding the yields and costs of these transformation rules, and ultimately can predict the most feasible retrosynthesis pathway.

Since retrosynthesis involves sequential truncations of the target molecule at various positions, the Monte Carlo tree search MCTS (namely an heuristic search algorithm for decision processes) is the technique of choice for making branch decisions. A 20-fold faster method than the traditional Monte Carlo method is the 3N-MCTS, that combines three different neural networks with MCTS to form a workflow for computer-assisted retrosynthesis planning.

But apart designing the routes of synthesis, AI algorithms can also effectively predict the products and yields of organic reactions on the basis of the molecular properties of the reactants. For example, quantum chemistry approaches — like the Hartree–Fock method,semi-empirical methods simplified versions of the Hartree-Fock theory anddensity functional theory (a computational quantum mechanical modelling) — can all potentially predict the products and yields of organic reactions, making the outcome of experiments efficiently modeled in silico.

In particular, several studies using AI algorithms to automatise, improve and generalise yield prediction have recently been published in this area. For example, in this study they demonstrated that ML can be used to predict the performance of a synthetic reaction in multidimensional chemical space using data obtained via high-throughput experimentation.

🎯 AI in automation of chemical synthesis⚗

Once a chemical reaction is planned, an holistic approach for the automation of chemical synthesis can follow. For example this can be done with the Chemputer platform operated by the Chempiler program — developed as a generalised standard which incorporates codified standard recipes, or chemical codes, for molecular synthesis — that accepts codified synthesis procedures from a scripting language called Chemical Assembly (ChASM). In the end, Chempiler is able to control the robotic operations so that users can directly run chemical synthesis reactions without manual reconfiguration.

This system had been validated by the successful synthesis of three pharmaceutical compounds: diphenhydramine hydrochloride, rufinamide and sildenafil, without any human intervention, and with yields and purities of products comparable with or better than those achieved manually.

Let's see now some data-driven startups involved in planning and execution of chemical synthesis.

🎯 AI startups involved in chemical synthesis ⚗

⚗Pending.ai (Wollongong, New South Wales, Australia 2018) utilises AI to learn chemistry from a database of more than 130 million compounds, 20 million reactions and 146,000 proteins, allowing researchers to generate novel molecules with neural networks, do structure-based drug design, plan chemical synthesis and conduct high-throughput chemistry. For example, Elsevier, a global research publishing and information analytics provider, is collaborating with Pending.AI to develop a predictive retrosynthesis tool based on deep learning.

⚗Molecule.one (Warsaw, Mazowieckie, Poland 2016) utilises AI to predict whether a particular chemical reaction will work. Molecule.one has raised a total of $385K💵 in funding over 2 rounds. Their latest funding was raised on Aug 19, 2019 from a Pre-Seed round.

⚗Chemical.ai is an AI company leveraging both human expertise and AI technology. They developed a customisable platform to plan synthetic routes in anything from a few seconds to a couple of minutes. Based on advanced ML and big-data technologies, their system displays the following characteristics: predicting the synthesis pathway for unreported molecules and finding new routes for reported molecules, sorting or filtering routes by cost, steps, green score, etc, finding more adaptable conditions for a clients’ reaction and potentially integrating clients’ Electronic Lab Notebook (ELN) data to provide reaction procedures. With customers leading CROs and big pharma, Chemical.AI provides one of the best CASP solutions in its sector. In March 2021, Chemical.AI announced the completion of a Series A funding of more than $5 Million💵, led by Sequoia China Seed Fund, and with participation from Nest.Bio Ventures and existing shareholder FreesFund.

⚗SciCalQ established in 2017, is a tech-based company providing AI solutions & computational chemistry services to academic or industrial organizations. They mainly focus on AI solutions in drug discovery, including the physical property predictions for lead compounds, ranking in drug repurposing, molecular dynamic simulations for large systems and Density functional theory (DFT) calculations. Chief Technology Officer Jianming Chen explains that clients would provide SciCalQ their data from synthesis that they can’t explain, and SciCalQ would use DFT calculations or another computational technique to help understand the mechanism underlying those data.

"We hope DFT service for chemists can be as convenient as scanning tunneling microscopy". — Jianming Chen, chief technology officer, SciCalQ

⚗ChemAlive sA (Lausanne, Vaud, Switzerland 2014) is an OpenSaaS automation environment focused on applying a mix of technologies, including quantum mechanics calculations and ML, for accurate prediction of chemical reactions and molecular properties, as well as modeling processes for drug discovery research. ChemAlive sA has raised 1 round. This was a Grant round raised on Nov 1, 2017.

⚗Hafnium Labs (Denmark 2018) is developing two software packages for high precision simulations of physical properties of pure components and mixtures. Their software products, Q-props and Epsilon, combine the latest advances in quantum chemistry, AI and cloud computing to predict chemistry to accelerate drug discovery, development of new materials, and processes. Hafnium Labs has raised a total of €1.6M💵 in funding over 2 rounds. Their latest funding was raised on Dec 1, 2019 from a Grant round.

⚗Zapata Computing (Boston, Massachusetts, United States 2017) is one of the key players in quantum computing technologies, that develops chemical simulations software for chemistry, logistics, finance, oil and gas industry, aviation, pharmaceuticals and materials. Zapata Computing has raised a total of $67.4M💵 in funding over 5 rounds. Their latest funding was raised on Dec 1, 2020 from a Venture - Series Unknown round.

⚗Qulab (Los Angeles, California, United States 2017) has built an integrated AI-powered drug design and chemical synthesis planning platform, named Quleap. The company claimed that once quantum computers become available, Qulab is positioned to become a front-runner in this area of the pharmaceutical research. Qulab has raised 2 rounds. Their latest funding was raised on Jan 1, 2021 from a Non-equity Assistance round.

Thank you for reading 👓💙

And if you liked this post why not share it?

@MetaphysicalCells

#science #health #pharma #drugdiscovery #drugdevelopment #AI #biotechAI

💻💊🔬💉📊🤖

👉🏻References

AI Drug Development Startups

Advancing Drug Discovery via Artificial Intelligence

AMPL: A Data-Driven Modeling Pipeline for Drug Discovery

Artificial intelligence in drug discovery (part 2)