Machine learning has outperformed human beings in several activities, including Go and chess, with global fanfare.
The technology has also demonstrated a strange ability to call unsuspecting people and hold a conversation with them.
However, a less-celebrated feat was DeepMind, the AI subsidiary of Alphabet, revealed at a biology conference that it has outperformed pharmaceutical companies in new drug discovery.
The initiative is also expected to build up the pressure for the leading global pharma companies to start preparing for a technological AI arms race.
Some new startups have already joined the race.
Back in December, at the Riviera Maya, Mexico-held CASP13 gathering, DeepMind outperformed experienced biologists in forecasting the structures of proteins.
This challenging pursuit has significant consequences: A device with the potential to model protein shapes accurately could help in expediting the development of new medicines.
“Absolutely stunning,” tweeted one scientist after the raw results were posted online. “It was a total surprise,” said conference founder John Moult, a University of Maryland computational biologist. “Compared to the history of what we had been able to do, it was pretty spectacular.”
Organizing the different protein shapes in an attempt to identify the ideal ways for medicines to fight illnesses is a considerable problem.
Researchers are yet to completely wrap their minds around the rules governing the process of creating proteins.
Worst case scenario, there’s a likelihood that protein structures are more compared to the atoms present in the world, which makes the process of prediction a daunting computation task.
For about 25 years, computational biologists have strived to come up with software that can get the task done.
Although Deepmind lacks vast experience when it comes to protein folding- how a protein obtains its 3D shape – it makes up for that with its newest neural network algorithms.
As such, the Alphabet-owned entity was able to accomplish what 50 of the leading laboratories in the world failed to do.
Excitement filled the Mexico-based resort, where the gathering took place.
Two presenters from DeepMind received numerous questions from scientists regarding how they managed to accomplish such a great feat.
Within hours of their presentation, The Guardian reported that DeepMind’s artificial intelligence could “usher in a new era of medical progress.”
Furthermore, the company boasted in a blogpost that how their protein models were “far more accurate than any that have come before,” opening up “new potential within drug discovery.”
In a brief email, DeepMind claimed that its researchers were “fully focused on their research” as opposed to being readily available to take part in interviews.
Nevertheless, Deep Mind’s simulation is yet to deliver the ideal atomic-type of resolution needed for the drug discovery process.
Even though numerous entities are striving to find new ways of using computers in identifying new medications, only a small number of machine learning-based medical drugs have received the go-ahead for testing in human beings.
It will take several years before anybody can find out whether such type of software has the potential to continually identify promising therapies that researchers may not have spotted before.
Artificial intelligence is undeniably an appealing catchphrase, especially in the health care sector today.
In fact, the technology is claimed to be the solution to all the problems affecting the industry.
DeepMind’s success could be an indication of a potential application for this revolutionary technology in one of the failure-prone and most expensive sections of the pharmaceutical enterprise.
Several observers claimed that the fact that outsiders could make such considerable progress in solving one of the most troubling biological issues is a challenge to all researchers in the particular field.
The effort by DeepMind could also be interpreted as a warning to the pharma industry, which is known for spending billions of money on R&D initiatives, but was still beaten by a tech company with no experience in drug discovery.
A Harvard-based computational biology specialist Mohammed AlQuraishi, who was present at the conference, reported in a blogpost that big pharma companies have failed to devote their attention to protein folding, ultimately leaving the space to technology companies.
While pharmaceutical companies were playing hesitant “Alphabet swoops in and sets up camp right in their backyard,” he wrote.
Discovering new drugs and availing them in the market is a daunting process.
Some estimates suggest that big pharmaceutical companies end up spending over $2.5 billion in a bid to get new drugs to patients who need them.
Only one out of the ten therapies that enter clinical trials for human beings successfully ends up in the pharmacy.
The science behind the drug discovery process is irrefutably slow.
In fact, in about two decades since the sequencing of the human genome, researchers have discovered cures for only a small portion of the nearly 7,000 identified rare illnesses.
What’s more, there are nearly 20,000 genes capable of malfunctioning in almost 100,000 ways, as well as millions of potential interactions that can happen between the ensuing proteins.
It’s difficult for drug hunters to probe such combinations without the help of technology.
“If we want to understand the other 97 percent of human biology, we will have to acknowledge it is too complex for humans,” claimed Chris Gibson, the co-founder, and CEO of Recursion Pharmaceuticals, which is a Salt Lake City-based company that leverages machine learning in hunting for new therapies.
Companies such as Recursion are increasingly attracting investors.
On the other hand, PitchBook, a data provider, revealed that VC firms invested $1.08 billion into machine learning and artificial intelligence startups with a focus on drug discovery in 2018.
The figure increased from a mere $237 million back in 2016, and the venture capitalists have already invested an additional $699 million so far in 2019.
On Monday, Recursion revealed that it raised about $121 million in its recent funding round, which involved the participation of various investors such as the Regents of the University of Minnesota and Intermountain Ventures.
According to PitchBook, the company is currently valued at $646 million.
“It is a very ambitious company. They are thinking in terms of radically changing the industry, “claimed Marina Record, an investment manager working for Scotland-based Baillie Gifford & Co, the company that led the financing round.
Established pharmaceutical companies are currently racing to form partnerships with entities that are undertaking the same work.
Back in April 2019, Gilead Sciences made deals with Insitro, a startup headed by the previous Stanford-based machine learning (ML) specialist Daphne Koller, which was aimed at finding ways of treating liver ailment NASH.
In the same month, AstraZeneca Plc joined forces with BenevolentAI in an attempt to find treatments for lung fibrosis and kidney disease.
Back in June, GlaxoSmithKline Plc collaborated with gene-editing experts based at the University of California in a $67 million partnership that will leverage artificial intelligence.
“Where else would you accept a 1-in-10 success rate?” said GlaxoSmithKline senior vice president Tony Wood, who heads medicinal science and technology for the British pharmaceutical giant. “If we could double that to 20% it would be phenomenal.”
Machine learning techniques “are going to be critical” to the drug discovery process, claimed Juan Alvarez, who is Merck Co’s associate VP for computational chemistry.
The giant pharmaceutical company is building artificial intelligence tools aimed at aiding its chemists in expediting the cumbersome process of coming up with chemicals for blocking aberrant proteins.
Early machine learning (ML) activities contributed in one way or another to drugs, particularly in human testing, even though the first medicines made using more sophisticated neural-network techniques could enter trials in a few years, said Alvarez.
AI could be utilized in scanning millions of cellular images – more than what human beings could ever do without technological help – in a bid to spot the therapies that could make cells affected by illnesses healthier.
At Recursion, which is among the earliest companies to utilize similar methods, robots apply thousands of possible medicines to different kinds of diseased cells, in about 400,000 – 500,000 small experiments that produce 5 -10 million images of cells.
Machine learning algorithms, on the other hand, begin scanning the particular images, to find the compounds that fight disease without harming any of the healthy cells.
Early algorithms were initially coded by using the hand in a bid to provide the interpretation for the basic cellular components.
However, Recursion is widely applying neural-network techniques that directly interpret the cellular images and may identify patterns that human programmers would not have been searching for in their line of work.
Computer scientists work together with biologists in the lab to come up with polished leads.
Recursion, which is currently engaged in rare-illness deals with Sanofi and Takeda Pharmaceuticals, produced over 2.5 petabytes of valuable data in the last few years.
What the startup is now doing “just wasn’t feasible six, or seven, or eight years ago,” claimed Gibson, its founder.
Gibson first used machine learning while studying at the University of Utah as a graduate student.
He was looking for cures for the cerebral cavernous disorder, which results in abnormal clumps of on permeable blood vessels located in the brain.
The Angioma Alliance revealed that the condition affects nearly one out of 500 individuals, and even though it is mostly silent, it can cause vision or speech difficulties, severe brain hemorrhages, and seizures.
Nearly a quarter of all patients suffer from a genetic form of this disorder, which is more likely to trigger numerous malformations.
Although the three keys genes behind this illness are well-known, there’s still no clinical treatment available.
One medicine Gibson tested while at the university was based on the existing knowledge of the illness caused severe symptoms in animals.
Annoyed with the results, Gibson and his team utilized open-source ML software to scan cellular images in an effort of probing the impacts of 2,100 compounds, while looking for the ones that boost the function and appearance of the blood vessel cells carrying the bad genes.
The algorithms revealed an unanticipated chemical that minimized leaky animal blood vessels by 50%.
The drug, which is expected to enter the second phase of human trials in 2020, led to the establishment of Recursion.
Other subsidiaries of Alphabet and the artificial intelligence research department of Facebook, which silently released a groundbreaking paper by using deep learning to assess 250 million protein sequences back in April, are gradually venturing into the pharmaceutical space.
In the spring of 2019, Google AI researchers revealed a neural network with the potential of forecasting a protein’s function from its amino acid sequence, which can assist biologists in comprehending what a new protein does.
Artificial intelligence supporters claim that no one is considering replacing human researchers with machines.
The objective is “augmenting and enhancing the decision-making capacity of scientists,” claimed Jackie Hunter, who is the GlaxoSmithKline’s former research executive and the current head of BenevolentAI’s clinical programs.
In the coming few years, there’s a likelihood that artificial intelligence-based simulations will be utilized in gauging whether potential medicines will be helpful before venturing into a comprehensive clinical trial.
An aerospace entity “won’t build and fly a plane without building it on the computer first and simulating it under many conditions,” asserted Colin Hill from GNS Healthcare, which is a startup that utilizes artificial intelligence (AI) in modeling illnesses.
Amgen.Inc makes ups one of the leading investors of GNS Healthcare.
In the coming years, pharmaceutical companies will not be in a position to commence clinical trials without conducting a virtual dry run, claimed Hill.
Nevertheless, what recently took place in Riviera Maya, Mexico has helped in expediting things.
DeepMind “basically beat everyone by a sizeable margin” stated AlQuraishi, a Harvard researcher.
He also added that in case drug makers fail to take the recent breakthrough as a severe threat, technological companies such as Alphabet’s DeepMind could outperform them.
Only time will tell whether pharmaceutical companies will start playing catch up or join forces with startups that are making impressive breakthroughs in the drug discovery space.