Finding the right talent is essential for AI – Dr Suneel Kumar, South Korea – INDIAai

  • Lauren
  • September 12, 2022
  • Comments Off on Finding the right talent is essential for AI – Dr Suneel Kumar, South Korea – INDIAai

Suneel Kumar is a computational chemist with expertise in computational chemistry, informatics, data analytics, deep learning and machine learning techniques.He has experience with integrated drug discovery partnerships and groups of international pharmaceutical and startup biotech clients.INDIAai interviewed Suneel to get his perspective on AI.Could you explain the role of AI/ML in Chemistry?The term ‘Machine Learning is not completely new to Chemistry. Both have long-time acquaintances. We have been doing “machine learning methods in chemistry” for two decades (traditional QSAR methods) to understand the structure-activity relationships and to develop regression and logistic models for property/toxicity predictions. With Big data, GPU-optimized deep learning algorithms allow us to work with data-driven modelling, generative models (SMILES/GRAPH), transfer learning, retrosynthesis prediction, multi-task predictions, and many more.,What piqued your interest in artificial intelligence?During my first assignment, I was exposed to the traditional statistical methods to understand/develop hypotheses to understand the structure and property relationship (we call it ‘QSAR’). It’s fascinating to apply those learning to design better drug-like molecules. That inspires me to explore more about AI/ML and its applications for drug discovery. So I have constantly been updating my skills in big data, machine learning, and deep learning approaches and trying those learnings to drug discovery.What are the emerging AI breakthroughs in drug discovery?So far, AI/ML has made significant contributions to drug discovery and development. Some scientific breakthroughs are AlphaFold (a database of predicted structures of 200 million proteins), GENTRL (and other generative methods) denovo generation of novel molecules based on the existing active and inactive compounds. A few more notable contributions are Deep Docking and Active Learning-guided docking approaches that improve the accuracy in predicting binding affinity and address false positives. In addition, graph-based and transformer-based methods proved to improve the accuracy in ADME/T endpoints.Is AI/ML in small-molecule drug discovery making any progress? Any notable contributions/success stories?As per the recent Nature report, 20 major AI-native companies have 160+ disclosed discovery programmes, and preclinical assets are in the pipeline. For example, Exscientia recently reported the phase 1 status of its AI-generated compound EX-21546 for cancer therapeutics. Exscientia identified its AI-generated compound (EX-21546) in 9 months by screening around 163 compounds. Schrodinger also reported its AI/ML driven clinical candidate SGR-1505 for blood cancer therapeutics; it identified this compound in 10 months by screening 78 compounds using its ML-powered platform. Not but not least, Insilico medicine also reported the Phase 1 status of its AI-designed anti-fibrotic drug candidate within 30 months (initialization to phase 1). Most traditional methods take around 4-5 years and need to screen thousands of compounds to identify the potential lead candidate with decent preclinical data. The programs mentioned above proved that implementing AI/ML approaches improved the success rate in identifying potential clinical candidates by screening fewer compounds within a short time. I hope we can reciprocate these learnings/approaches to more drug discovery programs and address the needs of the patients.The widespread misconception is that investing in AI infrastructure is expensive. Is this a myth?There is no correct answer for this. AI infrastructure cost depends on many factors. Size of the company, the amount of data, software building, algorithms of choice, and complexity of the AI solutions. As per recent statistics, most companies are roughly investing between $6,000 – $300,000 to set up AI infrastructure. In addition to the dedicated internal infrastructure, many cloud providers (AWS, Google, Azure, and many private players) offer cost-effective options.What are the critical factors in achieving effective AI transformation in the pharmaceutical industry?Here are some of my thoughts on key factors that are critical in AI transformation in the pharma industry:Data: As Andrew Ng states, “Data is food for AI” and explains the importance of high-quality data. Most data scientists spend ~80% of their time collecting, cleaning, labelling, and organizing the data for AI. Therefore, consistency and quality of the data are very much critical in data-centric modelling. As legends say, your ‘AI/ML model is only as good as your data’ – remember “, Garbage in, garbage out”. So make sure your data is reliable and diverse enough for feature learning. Try to filter out (duplicates, bad labels, and activity data points), revise twice and make sure the dataset is cleaned, labelled, ready and good enough for modelling.Right talent: Finding the right talent is critical in establishing AI in any industry. Right talent with domain expertise will understand the requirements and ensure to meet organizational objectives and the timelines of the deliverables.Explainable AI (XAi): Explainable AI models (XAi) are very much needed and critical for decision-making in drug discovery projects. It understands the model predictions, its false positives and false negative predictions. Furthermore, model predictability terms will help us to drive the drug discovery projects on the right path. So to conclude, there are many additional factors (such as. timelines, infrastructure, resources, data, and objectives) will help in the effective implementation of AI in the drug discovery industry.What are your responsibilities as PharmCADD’s Director of Computational Chemistry & AI?My role and responsibilities include:Evaluating AI-driven molecular design strategies (ligand and structure-based).Developing ExplainableAI/ML models.Contributing to internal drug discovery projects.My typical day starts with following the latest news articles, trends and research articles in deep learning from scientific journals and social media platforms. Then, I assess the new strategies and apply those learnings to live drug discovery projects and the project’s success.What advice do you have for individuals who aspire to pursue careers in AI research? What are the best paths to advancement?Here is my advice for scholars exploring AI/ML as a career:Find out the domain of your interest and identify the key areas that you would like to explore.Do an in-depth assessment of the critical areas, understand the current research and progress, and make a proper strategy based on the learnings.Actively follow the latest trends, articles, and workshops in your domain.Discuss your ideas and latest works with your fellow mates and seniors and try to understand their viewpointTry to follow top researchers in AI/ML on social media platforms (Youtube, Twitter, Medium and LinkedIn), follow their research and trends, and update yourself.Try to make a group of active researchers and participate in Hackathons and Kaggle competitions. Could you provide a list of notable academic books and publications?Machine Learning; Andrew Ng’s Coursera courseNatural Language Processing, Dan JurafskyAI/ML Book collections by Jason Brownlee Artificial Intelligence in Drug Discovery (ISSN) 1st Edition by Nathan Brown (Editor)Also, suggest you follow leading Youtube channels in data science (such as., KrishNaik, CodeEmporium, and codebasics) and medium blogs on deep learning. Also, follow active AI/ML researchers on Twitter and Linkedin to keep you updated with the latest updates in AI/ML. Also, follow essential workshops (such as., NeurIPS, ICML, ACL, etc.).