AI and the Lab of the Future

AI and the Lab of the Future In 2023, AI technologies broke into the public consciousness. But for lab leaders, AI tools have been growing in importance over the last decade. In the lab of the future, AI is likely to take an even more significant role in day-to-day processes. A September 2023 survey by the nonprofit Pistoia Alliance + looked at the lab technologies that life science companies intended to invest in over the next two years c What is AI? AI and machine learning (ML) came out on top, ranking as one AI has become a red-hot of the top three investment priorities for 60% of companies. buzzword across science. At its core, AI is a general term that covers a range of more specifi 60% techniques that leverage computing power to solve data problems. Let’s take a closer 51% look: Machine Learning (ML) 36% Uses algorithms that are “trained” to recognize patterns 29% and themes in datasets. This enables the algorithms to make inferences and predictions about future data they are presented with t Deep Learning A subset of ML that feeds data through structured, layered LIMS Cloud Data AI/ML Robotics/ networks called convolutional Plataforms Automation neural networks (CNNs). This facilitates more accurate Many other highly ranked technologies in the survey, analysis of complex systems. 4 such as robotics and cloud computing platforms, will come to incorporate AI over the next few years. Natural Language Processing (NLP) A digital approach to studying text that aims to automatically process linguistic elements in the text Machine Vision While NLP uses AI techniques to analyze text, machine vision studies imaging data. This involves image processing techniques like filtering pixel counting and pattern recognition The market is responding to these changing priorities – a n recent McKinsey research report identified almost 270 companies working on AI technologies in drug discovery .2 What Does an AI-Powered Lab Look Like? AI tools will be at the heart of the lab of the future. The Lawrence Berkeley National Laboratory’s A-Lab is a powerful demonstration of how AI could be used. A-Lab is a materials lab, carrying out assays on samples that could be useful in engineering applications such as solar and fuel cell development. s + The robot arms at A-Lab The robot picks from can together process 50– nearly 200 potential 100 times as many samples precursor molecules. as a human. A-Lab by the numbers: Working with a database of 150,000 new materials Robotic arms: 3 Furnaces: 8 Powder precursors: ~200 Lab size: 600 square feet Operates: 24/7 Samples tested per day: Data from the lab is 100–200 The lab’s central decision automatically sent to data making hub is controlled by repositories such as the AI software. Materials Project h AI in Drug Discovery The field of drug discovery has readily embraced AI. Tools powered by this software can help narrow down the millions of potential targets for small-molecule and biologic drugs. Target Screening + Assessing physiological targets for potential drugs involves AI Tools in Drug incorporating huge sources of omics data; AI can help this process Discovery12 f omics data; AI can help this process.9 AlphaFold 3D Protein structure prediction Drug Screening DeepChem A python-based library for ML tools like AlphaFold have allowed machine and deep learning researchers to predict the structure tools that can be used for of potential drug molecules tr molecular datasets DeepTox A pipeline for predicting chemical compounds' potential toxic effect Preclinical Screening RF-Score A drug's absorption and passage One of many tools that predicts through the body is a function of its binding affinity between physiochemical properties. These small molecule drugs and properties can be predicted using macromolecular targets ML methods to create safer and more effective compounds. a 4x 100,000 Sumitomo Dainippon Pharma worked with Insilico Medicine said that their Exscientia to identify DSP-1181 for obsessive compound INS018_055 is the first fully AI compulsive disorder 4 times faster than a typical generated drug. It targets a condition called drug development timeline – taking less than 12 idiopathic pulmonary fibrosis, which affect months vs an average of 4.5 years. 100,000 people in the US alone g l hi Lab-on-a-chip Lab-on-a-chip technology combines laboratory tools onto a small chip ne The technology works through microfluidics The chips are commonly linked to smartphones or computers where data analysis using AI can take place. The scale, throughput and cost-effectivenes that lab-on-a-chip designs enable can overcome one of the key challenges facing AI: generating appropriate data. The Challenges Facing AI in the Lab of the Future The Pistoia Alliance’s report also looked at what their life science leaders thought were the greatest barriers to their adoption of AI and other lab digitization. The graph below shows the percentage of respondents that identified issues as one of the top three obstacles. Unstructured data 58% Unstructured Data Even if data is available to train and power AI systems, it still has to be in an appropriate Data silos/no access data format for the AI to parse. Messy datasets, like 66% unclear handwritten notes, compromise the usefulness of tools like NLP. FAIRification of data 23% Data Silos/No Access to Data Inability to share/transfer analytical methods Data silos and access problems are common 14% within bulky company structures. Maximizing AI’s potential requires different sources of data to be made available for training Ontology management and analysis; cracking open these silos is essential work. 17% Lack of metadata standardization 42% Lack of Metadata Standardization AI systems are useful because they Lack of infrastructure and tooling analyze not just data, but also the data that 28% provides details about that data – metadata. By investing in informatics solutions like laboratory information management systems Resistance to data sharing/collaboration (LIMS), metadata can be organized and 48% mandated into lab practices. Lack of knowledge and expertize 28% Resistance to Data Sharing and Collaboration Lack of data governance within organizations Data systems can be upgraded overnight; 30% changing attitudes towards data sharing can be a generational task. Academics and researchers taught to preserve their IP Other and afraid of being scooped can block the 8% availability of data for AI systems. Sponsored by References www.technologynetworks.com/drug-discovery/articles/promises-and-challenges-of-target 1. Pistoia Alliance. Lab of the Future Report 2023.Pistoia Alliance. https://www.pistoiaalliance.org/ based- drug-discovery-355809. Published 2021. Accessed October 4, 2023. lab-of-the-future-report-2023/. Published 2023. Accessed October 4, 2023. 10. Lu M, Yin J, Zhu Q, et al. Artificial Intelligence in Pharmaceutical Sciences. Engineering. 2023. 2. Devereson A, Macak M, Nagra N, Idoux E, Stanzl E. AI in biopharma research: A time to focus and doi:10.1016/j.eng.2023.01.014. scale. McKinsey & Company. https://www.mckinsey.com/industries/life-sciences/our-insights/ 11. Rácz A, Bajusz D, Miranda-Quintana RA, Héberger K. Machine learning models for classificatio ai-in-biopharma-research-a-time-to-focus-and-scale. Published 2022. Accessed October 3, tasks related to drug safety. Mol Divers. 2021;25(3):1409-1424. doi:10.1007/s11030-021- 2023. 10239-x. 3. Sultana A, Maseera R, Rahamanulla A, Misiriya A. Emerging of artificial intelligence and 12. Sultana A, Maseera R, Rahamanulla A, Misiriya A. Emerging of artificial intelligence and technology in pharmaceuticals: review. Future J Pharm Sci. 2023;9(1):65. doi:10.1186/s43094- technology in pharmaceuticals: review. Future Journal of Pharmaceutical Sciences. 023-00517-w. 2023;9(1):65. doi:10.1186/s43094-023-00517-w. 4. Chassagnon G, Vakalopolou M, Paragios N, Revel MP. Deep learning: definition and perspectives 13. Sumitomo Pharma. Sumitomo Dainippon Pharma and Exscientia Joint Development New Drug for thoracic imaging. Eur Radiol. 2020;30(4):2021-2030. doi:10.1007/s00330-019-06564-3. Candidate Created Using Artificial Intelligence (AI) Begins Clinical Study. Sumitomo Pharma. 5. Liddy ED. Natural Language Processing. In: Drake M and Maack MN, eds. Encyclopedia of Library https://www.sumitomo-pharma.com/news/20200130.html#:~:text=DSP-1181%20was%20 and Information Science, 2nd. Ed. NY. Marcel Decker, Inc. 2001. created%20through,AI)%20platform%20for%20drug%20discovery. Published 2020. Accessed 6. Khan AI, Al-Habsi S. Machine Learning in Computer Vision. Procedia Comput Sci. October 4, 2023. 2020;167:1444-1451. doi:10.1016/j.procs.2020.03.355. 14. Insilico Medicine. First Drug Discovered and Designed with Generative AI Enters Phase II 7. Biron L. Meet the Autonomous Lab of the Future. Lawrence Berkeley National Laboratory. https:// Trials, with First Patients Dosed. Cision PR Newswire. https://www.prnewswire.com/apac/news newscenter.lbl.gov/2023/04/17/meet-the-autonomous-lab-of-the-future/. Published 2023. releases/ first-drug-discovered-and-designed-with-generative-ai-enters-phase-ii-trials-with Accessed October 4, 2023. first- patients-dosed-301862737.html. Published 2023. Accessed October 3, 2023. 8. Lawrence Berkeley National Laboratory. The Materials Project. Lawrence Berkeley National 15. Kuru CI, Ulucan-Karnak F, Akgöl S. 4 - Lab-on-a-chip sensors: recent trends and future Laboratory. https://next-gen.materialsproject.org/. Published 2022. Accessed October 4, 2023. applications. In: Barhoum A, Altintas Z, eds. Fundamentals of Sensor Technology. Sawston, UK. 9. Tay, A. Promises and Challenges of Target-Based Drug Discovery. Technology Networks. https:// Woodhead Publishing; 2023.