“Man cannot discover new oceans unless he has the courage to lose sight of the shore.” 📢 By Andre Gide
Latest studies on AI imaging and diagnosis 🧐
📌 A newly published study in the Lancet Oncology journal has found that the use of AI in mammogram cancer screening can safely cut radiologist workloads nearly in half without risk of increasing false-positive results.
📌 Researchers enabled ML to accurately predict Parkinson’s disease subtypes using stem cell images. This breakthrough showcases computer models classifying four Parkinson’s subtypes, with top accuracies reaching 95%.
📌 Preliminary results from a Swedish trial showed that AI is helping radiologists reduce their workload and detect more cases. In particular, scientists have found that AI-supported breast screening detected 20 per cent more cancers compared with the double reading of mammograms by two radiologists.
📌 A new study, led by researchers from Harvard Medical School, the University of Copenhagen, VA Boston Healthcare System, Dana-Farber Cancer Institute, and the Harvard T.H. Chan School of Public Health, has demonstrated that an AI tool can accurately identify individuals who are most susceptible to pancreatic cancer up to three years prior to their actual diagnosis, based solely on their medical records.
📌 Researchers found that AI outperformed most doctors on weekly diagnostic challenges, according to a study published in the medical journal JAMA (Accuracy of a Generative Artificial Intelligence Model in a Complex Diagnostic Challenge).
An NVIDIA leader shares his perspective on generative AI as a new prescription for success
Anthony Costa, the global lead of life sciences alliances at NVIDIA, just published an article about “What the Growth of Generative AI Means for Drug Discovery and Clinical Trials” discussing how the large language models (LLMs):
Can Exceed a Substandard Status Quo in Drug Discovery:
By Improving the speed and quality of early preclinical drug discovery pipelines, and even with small advancements in time-to-lead optimisation and improvements in the likelihood of clinical success.
Can Learn the Language of Biology:
Today generative AI models understand the language of biology, chemistry and genomic (that so far led to the advent of ChatGPT and other LLMs), although the demonstrated success of Google DeepMind’s AlphaFold was the seminal moment demonstrating the promise of these tools. Virtually every week, there are new LLMs published and made available, from academia to industry.
For example, DiffDock was recently released and generates a series of potential poses for protein-ligand binding, an approach that could lead to dramatic changes in the traditional drug development pipeline.
Other generative models trained on small molecules and protein data provide tools to virtually generate drug candidates with defined properties, and in a recent joint paper between NVIDIA and Evozyne ProT-VAE: Protein Transformer Variational AutoEncoder for Functional Protein Design have been synthesised and validated in the lab.
AlphaFold and ESMFold supported by evolutionary scale models in protein design, are the highest-quality property prediction tools yet known in the field, and directly contribute to the performance of structure prediction tools.
Applications of similar models to genomics have led to the first true generalisable foundation model for tasks such as gene expression prediction, were recently published by NVIDIA, the Technical University of Munich and InstaDeep acquired this year by BioNTech: The Nucleotide Transformer: Building and Evaluating Robust Foundation Models for Human Genomics.
Can Be a Prescription for Healthcare Success:
Other tools that integrate multimodal patient data, electronic health records, genomic and other representations of human health and biology are being built today to more effectively and efficiently recruit patients into clinical trials and to help them find trials themselves (NYU, NVIDIA Collaborate on Large Language Model to Predict Patient Readmission).
Another example, comes from Google and DeepMind that just announced Med-PaLM M, the first-ever Generalist Medical AI system, to provide high-quality answers to medical questions. Med-PaLM M (MPM) is based on PaLM-E, Google's robot model that combines language and vision, and is not a multimodal evolution of Med-PaLM 2, Google's large language model that has been refined for medical tasks. This makes sense since Med-PaLM M is supposed to be able to make diagnoses from visual data.
In particular, Med-PaLM M is the first multimodal Generative AI model that can perform a diverse range of biomedical tasks like: Classify medical images, Answer questions, Generate reports and Process genomes for diagnosis. And the implications of these models are just massive:
By providing doctors fast access to expertise across medical domains and modalities to improve decision-making.
By enabling rapid analysis of multimodal patient data for precision diagnosis and treatment.
By autonomously generate hypotheses combining insights across various disciplines, allowing medical discovery to explode. And
By democratising global access to medical expertise by allowing generalist models to be deployed widely.
AI: a paradigm shift in drug discovery by Dr. Kirsten Bischof and Dr. Gaurav Chandra (By The Yuan, a new community and technology platform dedicated to AI, ML, DL and related disciplines)
In their latest analysis, Dr. Kirsten Bischof (a trauma surgeon and critical care specialist in South Africa) and Dr. Gaurav Chandra (the chief operating officer of Enzolytics) discuss the challenges of some of the applications of AI for advanced research and drug development to address pharmaceutical problems, and give an overview of some notable breakthroughs in AI drug discovery so far. For example:
Insilico Medicine applied the generative adversarial network-based system GENTRL to rapidly identify potent discoidin domain receptor 1 kinase inhibitors within just 21 days. And now this First drug created by AI enters clinical trials.
DeepMind's AlfaFold learned to predict a protein's 3D shape from its amino-acid sequence, solving a 50-year-old challenge in biology. Read also the Top 5 Use Cases of AlphaFold in Life Sciences.
The University of Washington developed a DL model called RoseTTAFold that can calculate protein structures on a single gaming computer within 10 minutes (Efficient and accurate prediction of protein structure using RoseTTAFold2).
Canadian-based Deep Genomics made headlines in 2019 with discovering a novel target and a novel RNA therapeutics candidate for rare Wilson disease using their AI Workbench platform, all within 18 months of initiating target discovery effort (Deep Genomics Nominates Industry’s First AI-Discovered Therapeutic Candidate).
Enzolytics, used its Comprehensive AI protocol (for genomic surveillance and monitoring of viral epidemiology) and identified conserved targets for the monoclonal antibodies is producing (Enzolytics Announces the Discovery of Conserved Target Sites on the Monkeypox Virus).
Unlocking the potential of AI in drug discovery by wellcome
A new report by Wellcome a global charitable foundation established in 1936, explored the application AI in drug discovery, through a review of literature, patents, funding sources and a survey and interviews of practitioners, describing the current status, barriers and future opportunities for AI in drug discovery. This report is for researchers both in academia and in pharmaceutical/biotechnology industry, for funders and for policymakers, offering:
An overview of major applications and use cases of AI in drug discovery.
An assessment of the maturity of these use cases for different modalities (for example, small molecule drugs, biologics, vaccines), different stages of the drug discovery process and different therapeutic areas.
A review of the current adoption of AI in drug discovery, as well as barriers limiting its use. And
Suggestions on how to overcome barriers to unlock the potential of AI in drug discovery.
One of the key findings that emerges considers the barriers that must be addressed to unlock the full potential of AI, such as
Lack of trust in the value of AI.
Limited access, low maturity and lack of standardisation of data, tools and capabilities.
Market failure that may limit the applicability of key tools in less commercially attractive therapeutic areas – with the potential for AI to amplify disparities in health equity. And
Access to interdisciplinary capabilities such as computational chemistry and bioinformatics that has emerged as an additional barrier in many settings.
Always according the authors of this report, even though several initiatives are emerging to tackle all these barriers, however, on the current trajectory all these initiatives will be insufficient to unlock the potential of AI in drug discovery in ways that can equitably address urgent health needs. So, a solution for now will be to take these six key actions:
Find value from AI today.
Take no-regret moves to maximise future value.
Build coalitions to set 'rules of the road'.
Invest where AI intersects with drug discovery goals.
Contribute to the public debate.
Build organisational capabilities to enable delivery.
Latest news 📰
👉 Parexel, a clinical research organisation, and Partex, a biopharmaceutical company that has developed a digital pharma platform, announced a strategic alliance on Aug. 15, 2023, to leverage AI-powered solutions to accelerate drug discovery and development combining Parexel’s experience in Phase I to IV clinical development and Partex’s big data and AI capabilities (Parexel and Partex Enter AI Alliance).
👉 South Korea-based startup Standigm, a company using AI for drug discovery and development, has announced a collaboration with Nashville Biosciences, a wholly-owned subsidiary of Vanderbilt University Medical Center in the US, in order to accelerate early-stage drug discovery. Standigm will use Nashville Bioscience's extensive, de-identified genomic and clinical data to build new AI models for drug discovery (Korea-US partnership to revolutionise AI-based early drug discovery).
👉 The ex Google CEO Eric Schmidt wants his new nonprofit to become a big draw for top talent in science and AI (Ex-Google CEO Eric Schmidt to launch AI-science moonshot), and to potentially create breakthroughs in everything from drug discovery to material sciences.
👉 Just last month, Nvidia announced availability of DGX Cloud on Oracle Cloud Infrastructure for generative AI training. DGX Cloud, the cloud-based AI supercomputing service offering access to thousands of virtual Nvidia GPUs on Oracle Cloud Infrastructure along with infrastructure for training advanced models in generative AI and other fields, is already used by Amgen.
Amgen is using the DGX Cloud in combination with Nvidia BioNeMo large language model software and Nvidia AI Enterprise software, including Nvidia RAPIDS data science acceleration libraries.
👉 On August 10, 2023, Intelligent Omics Ltd (Intellomx) announced that it has joined Johnson & Johnson Innovation – JLABS, a premier life science incubator program. JLABS provides entrepreneurs with the lab space, resources and support needed to bring breakthrough healthcare solutions to patients around the world (AI Drug Discovery Pioneer Intellomx Joins Johnson & Johnson Innovation).
Intellomx is a leading UK-based drug discovery company that identifies with its Pathway Discovery Service the key drivers in disease pathways, revealing the underlying systems biology of multiple diseases. Their Digital Twin service provides a platform for rapid testing of off-target effects via disease pathways, saving costs and unnecessary animal trials. And their Novel Drug Discovery service uses proprietary artificial neural network algorithms to model non-linear biology, finding patterns that could not previously have been identified. They offer also the following services: Intuitive Intelligence, Novel Outcomes for New Drugs and Companion Diagnostics.
👉 The Top 6 Companies Using AI In Drug Discovery And Development according to Dr. Bertalan Mesko, PhD (The Medical Futurist and the Director of The Medical Futurist Institute, analysing how science fiction technologies can become a reality in medicine and healthcare) are:
The company that mines clinical trials: Antidote, focusing on matching patients and medical researchers in clinical trials, by combining proprietary technologies, data, and well-established business models. The company was launched under the name of TrialReach in 2010, but it got rebranded to Antidote in 2016.
The company with tangible results: Atomwise is the most well-known company in drug discovery using supercomputers to predict from a database of molecular structures which potential medicines will work, and which won’t. Their deep convolutional neural network, AtomNet, screens more than 100 million compounds each day.
The company that concentrates on cancer drugs: Turbine.AI has an AI solution to design personalised treatments for any cancer type or patient faster than any traditional healthcare service. Their technology models cell biology on the molecular level and can identify the best drug to target a specific tumour with; moreover can identify complex biomarkers and design combination therapies by performing millions of simulated experiments each day.
The company with a comprehensive structure from genomic to clinical data: Row Analytics specialises in digital health, precision medicine, genomics and semantic search. Their platform combines AI methods and data analytics to look at multiple genetic variants in combinations across a range of diseases. As they are able to complete the process in weeks instead of months, even for large disease populations with tens of thousands of patients, this enables the rapid identification of novel drug candidates and potential drugs to repurpose.
The company with research excellence in in-silico genomics: Deep Genomics, promises to solve the biggest puzzle in genetics: to get to know exactly what information the genome could provide for patients, medical professionals and researchers. For doing so, Deep Genomics is leveraging deep learning to help decode the meaning of the genome. So far, the company has used its computational system to develop a database that provides predictions for more than 300 million genetic variations that could affect the genetic code. For this reason, their findings are used for genome-based therapeutic development, molecular diagnostics, targeting biomarker discovery and assessing risks for genetic disorders.
The company with drug discovery in 46 days: Insilico Medicine, a top AI company for drug discovery, biomarker development and ageing research, that aims to cover the entire process of drug discovery, clinical trials analysis, and digital medicine using AI. It is pursuing internal drug discovery programs in cancer, dermatological diseases, fibrosis, Parkinson’s Disease, Alzheimer’s Disease, ALS, diabetes, sarcopenia and ageing.
👉 Last month, Recursion a leading clinical stage TechBio company decoding biology to industrialise drug discovery, launched Valence Labs that will serve as a ML research laboratory focused on developing the next generation of cutting-edge methods and models for drug discovery, and consists of the emerging ML research teams at Recursion and the team at Valence Discovery, which Recursion recently acquired (Recursion Launches Valence Labs at ICML with a Commitment to Open Science Including $1 Million in Academic Scholarships).
In particular, Recursion this year entered into agreements to acquire Cyclica and Valence, two companies in the AI-enabled drug discovery space. Subsequently, Recursion announced a collaboration and $50 million investment from NVIDIA to accelerate groundbreaking foundation models in AI-Enabled Drug Discovery. Then Recursion just launched Valence Labs, formerly Valence Discovery, a company with roots at Mila (a community in Quebec Canada of more than 1,000 researchers specialising in ML and dedicated to scientific excellence and innovation) and mentorship from Yoshua Bengio (a Canadian computer scientist, most noted for his work on AI networks and DL), dedicated to advancing DL in drug discovery, delivering impactful research and transformative technology and embracing open-source and open-science knowledge sharing with the machine learning community.
Finally, just this month Recursion announced an exciting milestone by successfully predicting the protein target interactions for approximately 36 billion chemical compounds in the Enamine REAL Space library, taking an important step to bridge the gap between the protein universe and the chemical universe (A Deep Dive into Screening 36 Billion Compounds). This was achieved using NVIDIA’s DGX Cloud supercomputing power and NVIDIA’s based supercomputer, BioHive-1.
👉 Talus Bioscience is a drug discovery and development company based in Seattle that is aiming to create new treatments for cancer, inflammation and other diseases. Recently, it received three new grants totalling $4.3 million to fuel its research for discovering new transcription factor inhibitors for two childhood cancers and accelerate the development of transcription factor drugs (Talus Bio Secures $4.3M in Grants to Fund Drug Discovery).
Talus Bio’s Multiplexed Assays for the Rational Modulation Of Transcription Factors platform, or MARMOT platform, work to disrupt transcription factors to stop disease. In particular, their platform measures transcription factors in live cells, where these proteins fold and function natively and this enable them to measure proteins in their natural, interactive molecular environments.
Then, the platform systematically measures all possible proteins in the nucleus, simultaneously, across diverse tissues, by using a unique team of experts in high-scale protein quantification. Accordingly, they have build ML models from the only database of unbiased transcription factor activity in the world they have, in order to predict new targets and new chemistry for undruggable targets. In the end, they have built a database of millions of compound-protein interactions from their growing internal compound database to discover privileged scaffolds for transcription factors.
👉 At last, Schrodinger's CEO Ramy Farid and Karen Akinsanya, Schrödinger’s President of R&D, Therapeutics, discuss in the following interview the company’s AI formula with GEN Edge: 🗣️ Schrödinger’s Equation: Physics + Machine Learning = Drug Discovery.
Schrödinger, is a company that stands out in drug discovery for combining AI with physics-based first principles to identify new drugs. And by applying physics this translates to running molecular dynamics simulations to compute the solubility of a molecule in water, or the affinity of the molecule for a particular protein, or its permeability.
“The calculations are slow, relatively speaking. It takes about a day to compute one property on one processor, approximately12–24 hours. And to do drug discovery, we need to explore hundreds of millions—billions, actually—of molecules. Even if you had one million computers, you couldn’t do that—and we don’t have access to one million computers!
So, we need this hack, if you will, to generate training sets, with physics that’s pretty fast to generate a large enough amount of data to train a machine-learned model. Machine-learned models are really fast, but you need the data to train them.
So there have been advances (with DL, neural networks, AI), but the fundamental limitation hasn’t changed.
You cannot build a training set in chemistry large enough because of the context. You have 1060 different molecules. You have 30,000 different proteins, and among each of those different proteins, you have different conformations and so on.
The complexity is way, way too high for machine learning alone.”
🗣️ Schrodinger's CEO Ramy Farid
Until next time 🌻🌞
PS: Webinars 📕📖👨🔬👩🔬 & Podcasts
📌 Discover the future of healthcare at Healthcare 2.0, the premier healthcare conference for industry leaders, innovators and entrepreneurs (Wednesday, September 27th 2023), to learn about
Generative AI based drug discovery vs clinical reality - will AI deliver on the clinical promise?
AI in Healthcare: an NHS perspective
AI impact across healthcare
📌 Join Drug Design Trends in the Age of AI by CDD Vault (Thursday, September 14, 2023), to learn about the intriguing world of computer-guided drug design, comparing two innovative methodologies:
fragment-based drug design and
full molecule-based drug design.
📌 Watch founder & CEO Philip Hemme at @FlotBio, Europe’s Biotech Podcast taking about #biotech