Focus Feature on Artificial Intelligence: 'Intelligence' in the discovery
AI is becoming far less of a buzzword and much more of a practical and necessary reality
Focus Feature: Artificial Intelligence
‘Intelligence’ in the discovery
AI is becoming far less a buzzword and much more of a practical and necessary reality
As in any industry, pharma and biotech are home to a lot of acronyms—many of them around new standards, techniques or technologies. For a while now, we’ve seen CRISPR snag a lot of headlines in the life sciences and in R&D for therapeutics, but that gene-editing technology has some competition from an acronym that has been around a lot longer: AI.
Of course, AI stands for artificial intelligence, and one of the reasons it hasn’t commanded more attention in academic labs or pharma R&D departments or in the news is that for decades, it has been largely theoretical. The idea of AI, the promise that it could hold—that was mostly what we had to work with.
Certainly, IBM’s Watson cognitive computing system began to change that thinking some years back when it beat a bunch of “Jeopardy!” contestants and then went on to a lot of other more substantive projects—setting us on a path of realizing that AI might actually be an achievable reality. But it’s only been recently that we’ve started to see a significant push by academia and industry to really use computing systems that employ machine learning and other tools at a high enough level to make AI actually come forth as a tool that people can use regularly and perhaps rely on—both of which are important, given the increasingly large and complex datasets people are working with in life sciences, pharma and biotech.
As we have been gathering information for this Focus Feature on the topic of artificial intelligence, the discovery phase of the R&D timeline really seems to be seeing a lot of action lately. For example, it was just at the beginning of January that San Francisco-based Numerate and Danish company Lundbeck Pharmaceutical announced a multitarget research collaboration to identify clinical candidates for the treatment of disorders in the central nervous system—including depression, psychosis, seizure and neurodegenerative disorders—using Numerate’s data-driven drug design process, which uses AI “to transform drug discovery.”
Aside from the potential to unlock new therapeutics, Numerate notes that the collaboration will increase the scale as it applies its AI-driven platform to drive the discovery of compound effectiveness and evaluate potential absorption, distribution, metabolism and excretion (ADME) and toxicity liabilities. Both partners tout how the deal “Further demonstrates the opportunity across the pharmaceutical industry in harnessing the power of AI to speed up drug discovery and development” and “overcome current drug design challenges to identify chemistries for ‘undruggable’ targets.”
“Utilizing data from Lundbeck combined with our internal modeling capabilities, we expect to produce multiple clinical candidates, while also continuing to refine, validate and expand our proprietary AI-driven platform,” said Guido Lanza, president and CEO of Numerate. “Importantly, neurology is an area where the ability to solve for the multiple, often competing objectives of activity, ADME (absorption, distribution, metabolism and excretion) and toxicity, is critical from the very beginning of a program. This multi-objective optimization of many causes of drug success and failure is the real power of applying AI to discovery. Specifically, our approach has allowed us to make significant breakthroughs in the prediction of ADME and toxicity. This creates a unique position for us to attack tough targets in the central nervous system (CNS) space when we combine it with our ability to model the real biology—the phenotype or functional effect of interest.”
And late last year, KemPharm Inc., a clinical-stage specialty pharmaceutical company focused on the discovery and development of proprietary prodrugs, and twoXAR Inc., an AI-driven biopharmaceutical company, announced that they had entered into a technology collaboration to develop prodrug-based therapies for multiple therapeutic areas and indications.
The collaboration will combine twoXAR’s AI-based technology to identify and de-risk drug product candidates, and KemPharm’s LAT (Ligand Activated Therapy) technology, which was developed to create new prodrugs that are designed to address unmet patient needs, improve the profile of drug product candidates and generate long-lived composition-of-matter patents.
Also in late 2018 came news out of Ireland, with Nuritas announcing the commercial launch of PeptAIde, which it is touting as “the world’s first bioactive ingredient discovered and delivered through artificial intelligence.” As the company explained, PeptAIde contains a patented network of anti-inflammatory bioactive peptides discovered by Nuritas’s proprietary AI platform, and is the first product from a deeper collaboration between Nuritas and BASF which aims to discover and commercially develop food-derived, natural bioactive peptides for health benefit.
“Harnessing the power of AI to significantly improve human health through new discovery is no longer a vision for the distant future,” said Emmet Browne, CEO of Nuritas, back in November. “Historically, AI has been associated with significant hype, but has yet to lead to the discovery of a healthcare product. With the launch of PeptAIde, Nuritas has delivered on that promise for the first time in human history by demonstrating AI’s true potential to improve health. AI has enabled us to do what would be previously considered impossible, propelling us from first partnership contract to commercial launch of a healthcare product in under two years.”
Innovation and investment
It’s not just about companies using AI more, or pharma/biotech teaming up with AI-focused companies. It’s also about an infrastructure that is quickly developing and evolving, just as when we saw a genomics infrastructure, increasingly affordable genetic sequencing technology, and then more accurate gene-editing tech, all in the wake of the Human Genome Project.
We see signs of that in the stories above, but also in news like that which came out in January, that Charles River Laboratories International Inc. and Atomwise Inc. had formed a strategic alliance to offer clients access to Atomwise’s AI-powered, structure-based, drug design technology, which allows scientists to predict how well a small molecule will bind to a target protein of interest.
As the new partners put it, “By removing sole reliance on empirical screening, AI enables drug researchers to test an extremely large and diverse chemical space in a matter of days and move through the optimization process quickly by focusing only on those compounds predicted to have improved target-binding attributes.”
And team-ups like this, much like we saw in genomics and then other omics-based advances, show us how important it is for the high-tech, next-gen technologies to integrate with other, more familiar technologies and pre-existing databases so that everything works seamlessly—at least, eventually.
In this case, the alliance combines two industry-leading drug discovery platforms: Atomwise’s AI technology and Charles River’s unique portfolio of end-to-end drug discovery and early-stage development capabilities and expertise. And by leveraging these platforms, there is the potential to significantly streamline the hit discovery, hit-to-lead and lead optimization process for clients’ research efforts.
“As Charles River continues to expand its early drug discovery portfolio, innovative solutions, including Atomwise’s AI technology, enable us to provide clients with a comprehensive, integrated platform for their early-stage drug research,” said James Foster, chairman, president and CEO of Charles River Laboratories. “By cutting time out of each stage of the drug discovery process, we enable our clients to deliver novel therapeutics to patients more efficiently and effectively.”
And late last year, we saw another sign of increasing interest and investment in AI with news from the United Kingdom that Optibrium, Intellegens and Medicines Discovery Catapult had secured a grant from Innovate UK to fund a £1-million project. That program will see Optibrium, which creates software to improve the efficiency and productivity of drug discovery, and Intellegens, a spin out from the University of Cambridge which is focused on a specific form of AI called deep learning, work with Medicines Discovery Catapult over the next two years.
As they note, “The aim is to harness the power of AI to learn from complex data and guide scientists in the design and testing of potential new drugs. Drug discovery generates a huge quantity of complex biological, chemical, clinical and safety information that needs to be collected, analyzed and presented in a way that it can be best used to make evidence-based decisions.”
The research partners are seeking a means of providing better insights into how a drug interacts with the body, improving the efficiency and productivity of drug discovery. The project will use novel deep-learning methods to create a next-generation platform that will better predict the ADME-toxicity of new drug candidates.
The work will build on Optibrium’s existing software offering, known as StarDrop, and an Intellegens’ deep-learning toolkit called Alchemite, and will develop and apply novel AI methods across ADME-Tox.
“We will apply cutting-edge deep-learning methods and new data to address important challenges in drug optimization,” vowed Matthew Segall, CEO of Optibrium. “The funding from Innovate UK is important validation of our project team’s expertise and the impact it will have on the industry’s efficiency and productivity. We look forward to making these models available to researchers through the StarDrop platform.”
Currently, high failure rates mean that nine out of 10 potential drugs fail somewhere between Phase 1 trials and regulatory approval, greatly adding to the costs of creating new medicines, the partners in this UK-based project note, adding: “The growth in computer-processing power, availability of large data sets, development of advanced algorithms and major improvements in machine learning presents a great opportunity for drug discovery. The technology and models created as a result of this project will help guide the selection and design of compounds early in the discovery process, to more quickly identify candidate drugs with a higher chance of success and lower development risk.”
Market potential and next steps
As GNS Healthcare noted in a post on its blog last year, “Hardly a day goes by without someone publishing an article on how artificial intelligence is revolutionizing the healthcare industry.”
The company acknowledged that AI is impacting multiple areas of the healthcare landscape, pointing to one recent survey that reported that 90 percent of pharma companies believe that AI is critical to their success, and another report that states that 42 percent of healthcare system leaders say they have or are planning to add AI as a tool for disease management.
But for all that, according to GNS Healthcare, the most important area in the life sciences and healthcare arena for AI to be might just be discovery, saying, “There is no question that AI is making an impact across the healthcare ecosystem from drug development to diagnostics to care interventions and physician tools to medication management. But among the wide-ranging applications of AI, one of the most important, and ultimately most beneficial, is its ability to accelerate the pace of discovery, helping healthcare stakeholders advance to the practice of precision medicine.
“The enormous power of AI, coupled with the increased availability of growing patient data sets and expanded computing power, is taking healthcare from population-level predictions of what may happen to novel discoveries of what will likely happen.”
Backing this up is a report out of Chicago from research intelligence company PreScouter in August of last year that provided a detailed look at the applications of AI in drug discovery and development, noting that the “use of AI in the pharmaceutical industry is projected to bring in billions of dollars in funding in the near future.”
“In the pharmaceutical industry, early use cases are becoming available that highlight the potential for AI to improve the process of discovering and developing a new drug, which is currently an incredibly difficult task,” said Dr. Charles Wright, PreScouter’s project architect for the healthcare and life-sciences industry.
Wright sees three common challenges that are faced by all pharmaceutical companies: (1) timelines of about 15 years; (2) costs in excess of $1 billion; and (3) a minuscule rate of success—mirroring the comments from Segall at Optibrium about high failure rate.
“It’s estimated that one in 10 small-molecule projects become candidates for clinical trials, [and] that’s after screening through millions of compounds to hone in on viable candidates,” noted Wright. “AI has the potential to transform the drug development process by making it both more efficient and effective, thus benefiting all parties involved—from the companies developing new drugs to the patients in desperate need of viable treatments.”
Dr. Navneeta Kaul, the second researcher who helped compile the report, takes it a step further, saying that he believes that with the advances made in AI, “The day is not far when a machine will be able to tailor a drug for each unique individual in a much shorter period of time.”
Looking back at what AI has done and where it might be going, GNS Healthcare noted, “Machine learning and AI have done a number of remarkable things, finding correlations and patterns in big data, and has worked well in the area of online advertising and retail. But in the world of healthcare and medicine, researchers, clinicians and patients need something beyond the correlations. Healthcare demands causality, because in order to know why, it is crucial to understand the underlying mechanisms of systems.
“The big breakthrough in AI that takes us from correlation to causality, and therefore provides the best means for discovering valuable new insights, is causal machine learning (CML). The most powerful form of machine learning used for AI, CML provides healthcare stakeholders with meaningful insights that simulate what will happen that results in better care management and more successful drug development.”
As we move through terms like AI, machine learning, CML, cognitive computing and more, there are going to be a lot more changes, refinements and additions. And, if what we saw in the wake of the Human Genome Project with gene sequencing and editing is any indication, we are going to see AI also advance very quickly. And that stands to be potentially very good news for drug discovery and development, which has labored under a very high-risk environment for a very long time.
Test uses AI to predict prostate cancer progression
NEW YORK—A pathology test called Precise MD that applies artificial intelligence (AI) to characterize tissue samples can accurately predict clinically significant prostate cancer disease progression following surgery, according to a study conducted at the Icahn School of Medicine at Mount Sinai and published in Nature Prostate Cancer and Prostatic Diseases.
Researchers at the Center for Computational and Systems Pathology at Mount Sinai used AI-guided machine-learning techniques to analyze cancer tissue samples from 590 patients who underwent a radical prostatectomy, an operation to remove the prostate gland and the tissues surrounding it. The Precise MD platform relies on cutting-edge microscopy with multispectral immunofluorescence to analyze cancer tissue architecture and biomarkers, enabling pathologists to see what the human eye cannot. Its analysis uses mathematical features to define tumor aggressiveness.
“By refining diagnoses, we can guide patients toward the best treatment option and optimize care,” said senior author Dr. Carlos Cordon-Cardo, chair of the Department of Pathology at the Mount Sinai Health System and professor of pathology, genetics and genomic sciences, and oncological sciences at the Icahn School of Medicine.
“Precision medicine is an innovative model of health care, and Mount Sinai is well positioned to provide our patients with more accurate diagnosis and tailored treatments,” said Dr. Dennis S. Charney, the Anne and Joel Ehrenkranz Dean of the Icahn School of Medicine at Mount Sinai, and president for academic affairs at Mount Sinai Health System. “Machine-learning systems in prostate cancer grading provide a more objective measure of risk assessment.”
The Henry Ford Hospital in Detroit and the Roswell Park Comprehensive Cancer Center in Buffalo, N.Y., provided patient samples for the study. Researchers say the Precise MD platform could be used to characterize any number of disease states, including but not limited to breast, melanoma, lung and colon cancers, as well as chronic inflammatory conditions such as inflammatory bowel disease.
Model ‘learns’ to make cancer treatment less toxic
CAMBRIDGE, Mass.—Massachusetts Institute of Technology (MIT) researchers are employing novel machine-learning techniques to improve the quality of life for patients by reducing toxic chemotherapy and radiotherapy dosing for glioblastoma, the most aggressive form of brain cancer.
In simulated trials of 50 patients, the machine-learning model designed treatment cycles that reduced the potency to a quarter or half of nearly all the doses, while maintaining the same tumor-shrinking potential. Many times, it skipped doses altogether, scheduling administrations only twice a year instead of monthly.
The researchers’ model uses a technique called reinforced learning (RL), a method inspired by behavioral psychology, in which a model learns to favor certain behavior that leads to a desired outcome.
According to MIT, the technique comprises artificially intelligent “agents” that complete “actions” in an unpredictable, complex environment to reach a desired “outcome.” Whenever it completes an action, the agent receives a “reward” or “penalty,” depending on whether or not the action works toward the outcome. Then, the agent adjusts its actions accordingly to achieve that outcome.
Rewards and penalties are basically positive and negative numbers, +1 or -1. Their values vary by the action taken, calculated by probability of succeeding or failing at the outcome, among other factors. The agent is essentially trying to numerically optimize all actions, based on reward and penalty values, to get to a maximum outcome score for a given task.
This approach was used to train the computer program DeepMind, which in 2016 made headlines for beating one of the world’s best human players in the game “Go.” It’s also used to train driverless cars in maneuvers such as merging into traffic or parking, which the vehicle will practice over and over, adjusting its course until it gets it right.
The researchers also designed the model to treat each patient individually, as well as in a single cohort, and achieved similar results (medical data for each patient was available to the researchers). Traditionally, a same dosing regimen is applied to groups of patients, but differences in tumor size, medical histories, genetic profiles and biomarkers can all change how a patient is treated. These variables are not considered during traditional clinical trial designs and other treatments, often leading to poor responses to therapy in large populations.
(To read the follow-up/sequel to this feature from our March issue, click here)