EVENTS | VIEW CALENDAR
Focus Feature: Artificial Intelligence 2.0
(This is a sequel/followup to the February Focus Feature on Artifiicial Intelligence)
Focus Feature: Artificial Intelligence 2.0
AI gives hope to patients and researchers
Artificial intelligence cannot solve every problem, but it has the potential to break up many bottlenecks
Human ingenuity has gotten us far over the centuries in terms of medicines. Technology has made the process faster and easier. And yet, at the same time, technology now gives us more than we can handle with our minds, particularly with voluminous genomic (and other omics) datasets and sometimes millions of leads to follow up. And that, of course, is where artificial intelligence (AI) is potentially a very great boon to life-sciences and therapeutic R&D.
Insilico Medicine, which focuses on artificial intelligence for drug discovery, biomarker development and aging research, is one of the many entities looking to capitalize on the hope of AI. One way it is doing that is with a research collaboration agreement announced last year with A2A Pharmaceuticals Inc., a biotechnology company headquartered in New York and focused on development of novel drugs for unmet needs in oncology, drug resistant bacterial infections and other life-threatening diseases.
What the two companies have done is create a fledgling company—Consortium.A—that will be tasked with applying the latest advances in AI to discovery of novel small molecules for rare and orphan disease, with a focus in particular at first on Duchenne muscular dystrophy (DMD).
As Insilico notes, computationally pre-optimized new drug candidates have already been designed for targets validated through its AI system. For its part, A2A Pharmaceuticals will assume the management of the new company, provide the development expertise for the newly discovered compounds and will serve as the contact point for any licensing of compounds.
“We are pleased to partner with Insilico Medicine, combining our strengths and complementary technologies to accelerate advancement of better therapeutics into the clinic for the patients that need them,” said Dr. Elena Diez Cecilia, head of business development at A2A. “Muscular dystrophy is a debilitating and terminal degenerative condition that causes muscle inflammation and wasting, and there is a huge need for more effective therapies.”
Both companies will collaborate on research programs devoted to the development of therapeutic approaches for DMD and other severe genetic disorders. Insilico Medicine’s technology applies advances in deep neural networks to identifying critical disease targets and generation of novel chemistry using next-generation artificial intelligence. A2A uses proprietary computational tools including artificial intelligence to design highly selective therapeutics for difficult-to-drug targets like protein-protein interactions.
“A2A Pharmaceuticals has a team of highly talented drug hunters with a proven track record in discovery, development and licensing of the drug candidates ... This is fantastic application for AI,” said Alex Zhavoronkov, founder and CEO of Insilico Medicine.
Head games for AI
On the not-so-rare disease side, Dr. Iya Khalil, co-founder and chief commercial officer of GNS Healthcare, talked recently on the company’s blog about another area of unmet—or at least incompletely met—need: migraines.
There are existing treatments for migraine and new ones that have come down the pipeline in recent years, but as Khalil points out, no one has been able to pin down the exact biological causes of migraines. There is a widely held belief is that the disease is a neurobiological disorder—an illness of the nervous system that’s caused by biological factors like genetics and metabolism.
Treatment options right now are generally twofold: pain-relieving medications and preventive medications. The FDA recently approved a drug that helps prevent migraines by targeting a protein called CGRP.
“Two obstacles stand in the way of progress however. First, the potential demand for new treatments may far outpace the supply of new drugs,” wrote Khalil. “The potential market for these drugs could be much larger than anticipated, as migraine often goes undiagnosed. Some reports estimate that 60 percent of women and 70 percent of men suffering from the condition have never been diagnosed with migraines.” The second challenge could come from insurance companies that might be hesitant to approve reimbursement, especially for expensive new preventive medications.
“Fortunately, more pharma companies are now leveraging AI and machine learning to help overcome these obstacles. This is particularly important as companies grapple with forecasting demand for newly developed therapies,” Khalil added. “Causal machine learning, a powerful form of AI, is poised to make a real impact in areas like migraine, where there is still much to understand about the disease. By developing causal disease models, biopharma is better able to analyze clinical trial results and extract value from data that often is inconclusive or riddled with confounding factors. These models can identify which patients respond to treatment and explain the crucial cause-and-effect relationships within the data. The result is often the identification of biomarkers for those patients who benefit in relation to the population as a whole.”
Speeding up the pipeline process
To even get to the point of clinical trials, however, you not only have to discover promising candidates, but optimize them and turn them into actual medications that can be administered in a trial setting. And that is one of the more time-consuming, as well as pricey, parts of drug R&D.
A recent article on the Massachusetts Institute of Technology (MIT) website mentioned that problem—how designing new molecules for pharmaceuticals is primarily a manual, time-consuming process that’s prone to error. But as that article also pointed out, MIT researchers have now taken a step toward fully automating the design process, which could drastically speed things up and produce better results.
As MIT explains, drug discovery relies on lead optimization, in which chemists take a promising lead (in the form of a molecule) and then tweak that chemical entity’s properties to get things like better uptake, higher potency, less toxicity and more.
“Even if they use systems that predict optimal chemical properties,” MIT noted, “chemists still need to do each modification step themselves. This can take hours for each iteration and may still not produce a valid drug candidate. Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Department of Electrical Engineering and Computer Science (EECS) have developed a model that better selects lead molecule candidates based on desired properties. It also modifies the molecular structure needed to achieve a higher potency, while ensuring the molecule is still chemically valid.”
“The motivation behind this was to replace the inefficient human modification process of designing molecules with automated iteration and assure the validity of the molecules we generate,” said Wengong Jin, a Ph.D. student in CSAIL and lead author of a paper describing the model.
The research was conducted as part of the Machine Learning for Pharmaceutical Discovery and Synthesis Consortium between MIT and eight pharmaceutical companies, announced in May. The consortium identified lead optimization as one key challenge in drug discovery.
“Today, it’s really a craft, which requires a lot of skilled chemists to succeed, and that’s what we want to improve,” said Regina Barzilay, the Delta Electronics Professor at CSAIL and EECS. “The next step is to take this technology from academia to use on real pharmaceutical design cases, and demonstrate that it can assist human chemists in doing their work, which can be challenging.”
Analyzing DNA data
There might not be any area more obviously in need of help from AI than the ever-growing mountains of genetic data we have accumulated since the Human Genome Project concluded and technological advances steadily reduced the difficulty and cost of DNA screening.
Many companies and academic institutions have been all over that problem for years, but there is always more to do. One of the more recent bits of movement in this realm comes out of Germany. Shivom, which offers a blockchain genomics platform for personalized healthcare, has partnered with Lifebit, an AI-powered DNA-analysis firm, to give users “unprecedented reporting capabilities for DNA data analysis.”
The partners claim that this collaboration will mean that immediate genome-wide association study analysis (GWAS) work will be possible without requiring specialist knowledge or an in-house data scientist.
Of course, we all know that specialist knowledge and scientists are always going to be needed at some stage, but much like technological advances have made it easier for scientists to use tools that once required highly specialized training, this work between Shivom and Lifebit might help democratize the ability to carry out GWAS efforts.
Also, as the partners note, it will mean that users “can access a library of pipelines [ready-built software used for analysis] and an AI-powered toolkit for analyzing the data in a way that is far more scalable than other solutions.”
Another aspect that they say makes the Shivom platform superior to other genomics platforms is that “[I]t will give pharmaceutical organizations and life-science users the ability to access real-time analysis whenever they need it with no waiting time, no application process for accessing the data and deadlines or cut off dates that restrict their access.”
“Through this partnership with Lifebit, we are providing enterprise users with the tools they need to find the right patients for their clinical trials more easily and more accurately than is possible through other solutions. Not only that, it brings AI into GWAS analysis in a way that hasn’t been seen before,” said Dr. Axel Schumacher, co-founder and chief scientific officer of Shivom.
The use of genomics platforms to improve rare disease treatment has increased in recent months, the two companies noted toward the end of last year when they made the announcement, with 23andMe partnering with GlaxoSmithKline to develop drugs for Parkinson’s disease. However, they say the Shivom-Lifebit partnership demonstrates a major leap forward in this area because it adds AI and machine learning capabilities to the identification of potential patients.
“Our partnership with Shivom will allow us to combine unique datasets with a level of analysis automation and insight generation that has never been seen before on a genomics platform,” said Dr. Maria Chatzou, co-founder and CEO of Lifebit. “In this way, scientists and doctors will be able to get all the benefits of this rich database without the need to rely on a data scientist for help. On the other hand, still ensuring that the individuals that have provided data to Shivom are given a level of security and control only a state-of-art blockchain technology can offer.”
AI for ‘sparse’ data
HELSINKI, Finland—Normally, we think of using AI for huge and complex data sets, but drug-enabling nanotechnology company Nanoform aims to enhance its proprietary STARMAP nanonization technology by applying “sparse data AI.” To that end, the company has announced Prof. Jukka Corander as head of artificial intelligence.
Corander is a world-leading expert in AI, Nanoform says, employing state-of-the-art machine learning techniques to create simulation-based models from sparse data. He is currently professor of biostatistics at the University of Oslo in Norway and professor of statistics at the University of Helsinki in Finland. His recent work with the Wellcome Sanger Institute Cambridge, U.K., includes the application of statistical machine learning and Bayesian inference algorithms on biological data. Corander will apply his expertise to further develop Nanoform’s STARMAP. The implementation of AI will help define the physical characteristics of drug candidate molecules from limited data to understand how these parameters influence solubility and bioavailability.
Sparse data AI will combine with Nanoform’s best-in-class technology to predict nanonization success for new drug candidates and form a more efficient particle engineering process for drug development. The software will also be used to enhance Nanoform’s manufacturing process by implementing deep learning for consistent, iterative improvement.
Nanoform’s appointment of a head of AI is, the company says, in response to the significant interest in AI for drug discovery, as “AI can be used to model alternative applications of current drug compounds and determine how particle engineering can produce optimal drug design and formulation. Nanoform’s partners are set to benefit from this innovative approach to drug discovery and development, which significantly increases the likelihood of identifying successful compounds that can quickly progress to market.”
AI system designs drugs from scratch
CHAPEL HILL, N.C.—An artificial-intelligence approach created at the University of North Carolina at Chapel Hill Eshelman School of Pharmacy reportedly can teach itself to design new drug molecules from scratch and has the potential to dramatically accelerate the design of new drug candidates, according to news out of the university in summer 2018.
The system is called Reinforcement Learning for Structural Evolution, known as ReLeaSE, and is an algorithm and computer program that comprises two neural networks which can be thought of as a teacher and a student. The teacher knows the syntax and linguistic rules behind the vocabulary of chemical structures for about 1.7 million known biologically active molecules. By working with the teacher, the student learns over time and becomes better at proposing molecules that are likely to be useful as new medicines.
“If we compare this process to learning a language, then after the student learns the molecular alphabet and the rules of the language, they can create new ‘words,’ or molecules,” said Alexander Tropsha of the UNC Eshelman School of Pharmacy and co-creator of ReLeaSE along with Olexandr Isayev and Mariya Popova. “If the new molecule is realistic and has the desired effect, the teacher approves. If not, the teacher disapproves, forcing the student to avoid bad molecules and create good ones.”
The University has applied for a patent for the technology, and the team published a proof-of-concept study in the journal Science Advances in July 2018.
According to the university, ReLeaSE is a new innovation to virtual screening, the computational method widely used by the pharmaceutical industry to identify viable drug candidates. But whereas virtual screening allows scientists to evaluate existing large chemical libraries for known chemicals, ReLeASE has the unique ability to create and evaluate new molecules.
AI-driven discovery company closes Series B round
OSFORD, U.K.—Exscientia, an artificial intelligence (AI)-driven drug discovery company, announced in January it had raised $26 million in a Series B financing round, which will be used to scale the company’s pipeline and advance selected programs toward clinical development.
One of the new investors was Celgene, and as noted by Dr. Rupert Vessey, president of research and early development there: “Exscientia has demonstrated that AI in molecular design is here today. With the global pharmaceutical industry acknowledging the importance of incorporating AI-driven R&D approaches into their drug discovery processes, we see a huge growth opportunity ahead. We believe Exscientia is set to become a global leader in AI-driven drug discovery and are excited to participate in this investment round.”
Exscientia has made what it calls “considerable progress during 2018” and anticipates its first programs driven by AI will be IND-ready by early 2020, if not before.
“This Series B marks a milestone in our development and enables us to drive the next phase of strong business growth. Over the past 12 months we have substantially expanded our operations and capabilities to become a full stack AI drug discovery company,” said Prof. Andrew Hopkins, CEO and founder of Exscientia. “Furthermore, our unique Centaur Chemist platform allows us to move rapidly from idea generation to new drug molecules ready for IND and clinical development. With this new funding Exscientia is positioned to become the dominant player in AI drug discovery, driving radical change in R&D productivity. We are excited Celgene and GT Healthcare have joined with existing investor Evotec on this exciting journey.”
Added Dr. Werner Lanthaler, CEO of Evotec: “We continue to be very impressed with the progress Exscientia has made over the past year. Through our partnership with Exscientia, we have seen first-hand evidence that they can deliver the most productive drug discovery engine in the industry. This latest funding will allow Exscientia to apply its platform at scale, taking advantage of the efficiencies that its AI-driven systems provide.”
Kaia Health unveils study to explore digital COPD treatment
NEW YORK—Digital therapeutics company Kaia Health, which uses AI-powered motion tracking technology to, as it says, “tackle some of the world’s most urgent health challenges,” recently unveiled a feasibility study to examine the impact of its digital therapeutic treatment of chronic obstructive pulmonary disease (COPD) in Japan’s aging population. This follows a German pilot study which successfully decreased symptoms.
The Kaia Health COPD app addresses physical and psychological factors of the disease, the company says, further adding that it is based on clinically validated patient guidelines and allows users to better self-manage their COPD. The app includes video-based physiotherapy which offers exercises to help patients build muscle and promote a healthy cardiovascular system, whilst a training algorithm adjusts the support based on each patient’s disease profile and feedback.
Psychosocial support is provided through audio-based relaxation exercises to manage anxiety and depression and to cope with dyspnea attacks. Patients can also contact a coach via the app who will answer app-specific questions, work with users on their individual goals and offer motivation.