Eye on AI - December 11th, 2020
Welcome to Aigora's "Eye on AI" series, where we round up exciting news at the intersection of consumer science and artificial intelligence!
This week, we’ll be taking a closer look at two potentially groundbreaking announcements in AI science: the first AI-synthesized drug to go to clinical trial, and Google DeepMind’s breakthrough in protein folding.
First AI-Synthesized Drug Goes to Clinical Trials, Marking Breakthrough in AI Medicine
We begin with truly revolutionary news in AI-synthesized medicine. According to the AIM blog article “Breakthrough: Drug Synthesized By AI Goes Into Clinical Trials,” the first precision-engineered drug generated by AI, an OCD medication developed by the UK-based Exscientia in collaboration with Japanese Pharma, Sumitomo Dainippon Pharma, has gone into clinical trials, marking what many are calling one of the most significant milestones in the history of artificial intelligence.
“Creating drugs is a process that can occupy researchers for many years,” writes AIM contributor Sohail Merchant. “However, utilizing AI, the discovery of the latest OCD medication took just 12 months instead of the usual 60 (or five years). Also, AI could locate the candidate compound (an antibody or small molecule with a strong therapeutic potential whose activity can be optimized to create drugs) within 350 synthesized compounds rather than the conventional 2,500.”
The molecule used in the new OCD drug is known as DSP-1181. It was created by utilizing algorithms to sift through potential compounds that would react beneficially against a massive database of OCD parameters, then selected compounds that might positively react to those set of parameters to identify optimal candidate molecules––in other words, the time it took to AI identify a potential drug solution for OCD was much faster than expected.
The new synthesized OCD drug is said to activate 5-HT1A receptor, decreasing aggression, and drug-seeking behavior yet increases sociability, prolonged REM Sleep, and heightened sex drive and arousal, all positive predictions for a successful trial. The drug is still untested, so it’s too early to jump the gun on success rates. That said, previous OCD drug treatments only showed improvement in 75% of patients, leaving plenty of room for improvement. The bigger picture is even more exciting. The use of AI and deep learning to design, synthesize, and validate novel drug candidates is fifteen times faster than the best pharma companies. If all goes well with the trial, we could see more AI-synthesized medicines used to cure diseases that were once thought incurable!
AlphaFold’s Breakthrough in Protein Folding is Bigger News Than You Think
Let’s continue with more big news for AI science. As news broke last month that Google’s AlphaFold 2 indisputably won the 14th Critical Assessment of Structural Prediction competition, a biannual blind test where computational biologists try to predict the structure of several proteins whose structure has been determined experimentally, many were left wondering what that actually meant for science. The Oxford Protein Informatics Group (OPIG) attempted to answer that question this week with their article, “CASP14: what Google DeepMind’s AlphaFold 2 really achieved, and what it means for protein folding, biology and bioinformatics.”
“Protein structure is at the core of biochemistry, and has profound implications for medicine and technology,” notes the OPIG. “Establishing the structure of a protein is a bottleneck in structure-based drug discovery, and accurate structure prediction is expected to improve the productivity of pharmaceutical research pipelines (although it is only one factor, and we will need to get other things right before truly revolutionary changes happen — check Derek Lowe’s posts here and here). Structural information of proteins is also essential in biology, where it helps to elucidate function — many key papers in biochemistry derive insight from experimental advances in structure determination.”
Protein folding has been one of those lasting biological conundrums that researchers have been struggling with for decades, which makes AlphaFold’s discovery even that more impressive. After the announcement, researchers were left questioning just how successful AlphaFold’s results were. The answer, according to OPIG, was very.
“.... after three decades of competitions, the assessors declared that AlphaFold 2 had succeeded in solving a challenge open for 50 years: to develop a method that can accurately, generally and competitively predict a protein structure from its sequence (or, well, a multiple sequence alignment, as we will see later),” continues the OPIG. “There are caveats and edge cases, as in any application — but the magnitude of the breakthrough, as well as its potential impact, are undeniable.”
As of now, the question of if and how DeepMind will make their code available still lingers. It’s currently being discussed at Google, with an expected announcement in January. If given over for communal use, the code would provide a general solution for protein structure prediction. It’s not universal. Several of the CASP14 targets were not predicted successfully, suggesting that there are some protein families that require further work. Even still, the implications for science are monumental.
“What Google just achieved might very well be among the most important scientific achievements this century, in terms of impact if not epistemologically,” continues the OPIG. “The long sought-after ability to predict the structure of a protein from its sequence (and, as of yet, availability of similar mutated sequences) will unlock applications spanning the entirety of the life and medical sciences, from basic biology to pharmaceutical applications. The prospects are truly astounding.”
For a much more detailed look than I can provide at how AlphaFold’s code works, and how it could affect scientific research, I suggest a good hard look at the article. This is truly groundbreaking stuff, and I’m eager to see whether Google releases the code for wider use.
That's it for now. If you'd like to receive email updates from Aigora, including weekly video recaps of our blog activity, click on the button below to join our email list. Thanks for stopping by!