Back to News

Algorithms to Treatments: AI’s Impact on Modern Drug DiscoveryAuthor: Dr. Somesh Sharma, EVP & Head, Discovery Services

Artificial Intelligence (AI) is revolutionizing drug discovery, shifting it from a lengthy and uncertain process to one marked by rapid precision and innovation in healthcare. By harnessing vast datasets and uncovering complex patterns, AI dramatically speeds up the development of new treatments, making groundbreaking medical advancements more attainable.

AI-Driven Drug Discovery: Pioneering the Future of Healthcare

Artificial Intelligence (AI) is transforming drug discovery by enhancing efficiency, reducing costs, and increasing success rates across various stages, including target identification, hit discovery, lead optimization, process development and manufacturing. AI technologies, such as knowledge graphs for OMICs data mining, Generative AI (Gen AI) for molecule design, and structure prediction algorithms like AlphaFold, are advancing drug development strategies. These technologies facilitate tasks like protein-ligand docking and virtual screening with unprecedented accuracy, building new chemical space and opening new therapeutic avenues.

From Concept to Cure: Advances in De Novo Drug Design

De novo drug design, which entails the creation of novel drug-like molecules ab initio, has been profoundly transformed by advancements in Generative AI. Contemporary methodologies encompass Simplified Molecular Input Line Entry System (SMILES) based models, which encode molecular structures as linear strings, and molecular graph-based models, which utilize three-dimensional representations to enhance structural fidelity and robustness.

Innovations such as DeepSMILES and SELFIES (Self-Referencing Embedded Strings) have improved the encoding and decoding processes, ensuring higher validity and diversity of generated molecules. Reinforcement Learning (RL) techniques, including Reinforcement Learning-Variational Autoencoder (RL-VAE), further augment these models by optimizing molecular properties through iterative feedback mechanisms.

Moreover, the integration of genetic algorithms with deep learning frameworks, exemplified by Genetic Algorithm with Reinforcement Learning (GARel) and Genetic Evolutionary Reinforcement Algorithm (GENERA), has refined de novo design. These hybrid approaches focus on the generation of unique molecular scaffolds and perform multi-objective optimizations, thereby yielding compounds with enhanced pharmacokinetic and pharmacodynamic properties.

AI-Powered Virtual Screening: Accelerating Therapeutic Discoveries

Traditional virtual screening methods predominantly utilized rigid docking simulations and empirical scoring functions, which constrained the exploration of chemical space. However, the advent of Gen AI, ML, and DL has revolutionized virtual screening in drug discovery, markedly improving the efficiency and accuracy of identifying potential drug candidates.

For instance, Generative Adversarial Networks (GANs), which employ generator-discriminator networks, and Variational Autoencoders (VAEs), which encode molecular data into latent spaces, have significantly broadened the chemical space explored during virtual screening. ML algorithms such as Random Forests (RFs), Support Vector Machines (SVMs), and k-Nearest Neighbors (KNN) analyze extensive datasets to predict drug-target interactions and classify compounds. Deep Learning techniques, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), model intricate data patterns to enhance prediction accuracy.

AI Enable Chemistry Platforms for Faster Deliveries

Integration of AI in chemistry laboratories is indeed transformative. AI-powered ‘self-driving labs’ can autonomously conduct experiments, analyze data, and integrate with biological tools. This significantly accelerates the Design-Make-Test-Analyze (DMTA) cycle, enabling faster decision-making and discovery of novel molecules.  AI tools are crucial in optimizing chemical processes. They help in minimizing waste generation and reducing energy consumption, aligning with the principles of ‘Green chemistry’, making the processes more efficient and environmentally friendly.

Advanced AI tools like SYNTHIA™, Chemical.AI, and ASKCOS assist chemists in identifying the most efficient and cost-effective synthetic routes. These tools can rapidly propose viable routes of synthesis, significantly shortening time required for synthesis. AI tools are also employed for process optimization, predicting impurities, and assessing genotoxicity. These capabilities are essential for ensuring safety and efficacy of chemical processes and products.

Advancements in In Silico ADMET Modelling

AI has significantly impacted all areas of drug discovery research, with its influence being particularly profound in in silico ADME modelling, was primarily driven by high clinical trial failure rates in the late 1990s due to poor pharmacokinetics. Initially, property-based drug-likeness rules and high-throughput assays were adopted for early evaluations of drug efficacy and safety.

Machine Learning (ML) algorithms, including Random Forests (RFs), Support Vector Machines (SVMs), and k-Nearest Neighbors (KNN), along with Deep Learning (DL) techniques such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), have further advanced ADMET predictions by modeling complex patterns. Graph Neural Networks (GNNs) optimize chemical structures by modeling interactions between molecules and their targets. Tools such as ADMET-AI and ChemMORT exemplify these advancements, offering rapid analysis and enhanced lead optimization.

In the domain of drug toxicity prediction, AI and DL models have significantly advanced the field by analyzing extensive datasets to forecast potential adverse effects. Tools like DeepTox utilize molecular descriptors to predict toxicity, while Deep-PK estimates both toxicity and pharmacokinetics, enabling earlier and more accurate identification of potential drug-related issues.

Gen AI, employing algorithms such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), advances drug discovery by designing novel molecular structures. VAEs encode and decode molecular data to explore diverse chemical spaces and propose innovative drug candidates.

AI Impact on In vitro Assay Development

AI is playing a pivotal role in the development of in vitro assays, enhancing their efficiency, accuracy, and reproducibility. AI algorithms can design and optimize assays by analysing large datasets to identify the most relevant parameters and conditions for assay robustness. AI tools can process and interpret complex data from in vitro assays, identifying patterns and correlations for a meaningful information, which might lead to more reliable and reproducible assay results. 

AI-powered applications can analyse vast amount of data generated through high-throughput screening of compounds in a fraction of the time and enhances the speed and efficiency of screening process.

 AI-Driven Drug Optimization

GANs and VAEs models are crucial for optimizing drug candidates by predicting how structural changes impact biological activity and safety. BenevolentAI uses these models to improve amyotrophic lateral sclerosis drug candidates, integrating knowledge graphs for better efficacy and safety. Exscientia’s Centaur Chemist™ AI platform, with Sumitomo Dainippon Pharma, identified a novel OCD drug candidate in 12 months, accelerating the process. Insilico Medicine’s Pharma.AI platform combines GANs and reinforcement learning to develop molecules for idiopathic pulmonary fibrosis, showing promising trial results. Atomwise’s AtomNet® employs deep learning for structure-based drug design, identifying Ebola virus inhibitors. DeepMind’s AlphaFold has revolutionized drug development by predicting protein 3D structures in minutes, with AlphaFold 4.0 enhancing accuracy. Meanwhile, Chai-1 by Chai Discovery sets a new benchmark in molecular structure prediction, surpassing AlphaFold on several benchmarks and offering adaptability for small molecules, proteins, DNA, RNA, and chemical modifications, making it a crucial tool for precise drug development.

AI-Designed Molecules Making Clinical Advances

There are several examples which illustrate the impact of AI in drug discovery with rapid identification and optimisation of novel drug candidates for clinical trials. A few prominent examples include EXS-21546, an A2A receptor antagonist developed by Exscientia and Evotec for treating solid tumors with high adenosine signatures, which is currently in Phase 1/2 clinical trials. DSP-1181, a 5-HT1a agonist for obsessive-compulsive disorder, developed through a collaboration between Exscientia and Sumitomo Dainippon Pharma, has entered Phase 1 clinical trials. Another example is EXS-4318, a selective protein kinase C-theta inhibitor, which is in Phase 1 clinical trials. INS018-055, a TRAF2- and NCK-interacting kinase inhibitor for idiopathic pulmonary fibrosis, developed by Insilico Medicine, has progressed to Phase 2 clinical trials. Additionally, RLY-4008, a highly selective FGFR2 inhibitor for cholangiocarcinoma and other solid tumors, is currently in Phase 1/2 clinical trials. These developments highlight the growing influence of AI in advancing drug discovery.

Challenges and Collaborations in AI-Driven Drug Discovery

AI-driven drug discovery faces multiple challenges, though, it has potential to revolutionise drug discovery. The primary challenge lies in obtaining high-quality data necessary for developing a robust predictive tool. The use of AI raises serious concern on ethical issues, such as data privacy, consent and potential for constructing biased algorithms. The other operational challenge could be in unifying existing workflows and database to integrated AI tools.

To address these challenges and opportunities require continuous refinement, engagement and strong collaborations. A partnership between pharmaceutical companies and tech firms such as Evotec and Exscientia, resulted in the development of EXS-21546, now in clinical trials for solid tumors. A collaborative initiative like the open consortium unite stakeholders to share resources and drive innovation. An evaluation from Critical Assessment of Computational Methods for Protein Structure Prediction (CACHE) and Critical Assessment of Structure Prediction (CASP) are essential initiatives for differentiating between the reality and hype surrounding AI in drug discovery. These assessments are crucial for validating the effectiveness of AI-driven tools and ensuring that they provide practical, reliable results rather than just theoretical or over-hyped claims.

Future Prospects of AI in Drug Discovery

Despite the inherent challenges, AI is poised to play an increasingly pivotal role in drug discovery. An enhanced collaboration and data sharing through open consortia initiatives will further propel innovation in this field. There is a growing emphasis on developing ethical and regulatory frameworks to address data privacy, algorithmic bias, and accountability. The shift from theoretical research to practical applications will broaden AI’s accessibility, significantly impacting therapeutic options and improved health outcomes. The future of AI in drug discovery promises substantial advancements, improving research efficiency and leading to more effective, personalized treatments

Source – Pharma Bio World – https://jasubhaimedia.com/pbw/2024/october-2024/#p=24