In the world of artificial intelligence and machine learning, Google Deepmind’s AlphaGo program defeating world champion Lee Sedol at the game Go on March 19, 2016, was a watershed moment. It showed, among other things, that if an AI algorithm is trained on a vast array of data points and constantly improves through iterations, it can defeat the most advanced neural network: the human brain.
A massive treasure trove of internally-developed dataset of chemical compounds that can be interrogated with AI is what upstart biotech startup, Anagenex, is promising to bring to the world of small-molecule drug discovery. By combining large scale lab experiments with computational tools, it hopes to upend the drug discovery game that requires infinite patience and even more greenbacks. The Boston-based company announced on Wednesday that it has raised $30 million in a Series A funding round to get closer to its ultimate goal of developing life-saving medication for previously undruggable targets. Catalio Capital Management led the round along with participation from Lux Capital, Khosla Ventures, Obvious Ventures, Airstreet Capital, and Menlo Ventures. A previous seed round brought in $7.2 million.
“We see a lot of platform technologies, but were blown away by Anagenex’s potential to fundamentally reshape how small molecules drugs are discovered,” said George Petrocheilos, general partner at Catalio in a news release announcing the fundraise. “Going after targets that have frustrated the industry for decades is always a risky business. The power and efficiency of Anagenex’s platform makes that risk tolerable, especially considering the potential payout.”
What is so powerful about the platform?
In a lengthy email response to questions, Anagenex’s CEO, Nicolas Tilmans contended that other companies that call themselves AI drug discovery firms have a fraction of the dataset to train their AI on. Further, they are dependent on external datasets, often from clinical research organizations that may be of poor quality. Tilmans declared that “even if they were of high quality, traditional drug discovery datasets are 100-1000x smaller than they need to be to leverage modern ML methods,” whereas Anagenex can “generate datasets that are the right size (billions+ of data points).”
Another key differentiator, he believes, is how the company tests an initial population of compounds and feeds back the information garnered from these tests back into the AI engine so that it learns from it.
A Deep-dive Into Specialty Pharma
A specialty drug is a class of prescription medications used to treat complex, chronic or rare medical conditions. Although this classification was originally intended to define the treatment of rare, also termed “orphan” diseases, affecting fewer than 200,000 people in the US, more recently, specialty drugs have emerged as the cornerstone of treatment for chronic and complex diseases such as cancer, autoimmune conditions, diabetes, hepatitis C, and HIV/AIDS.
“We can iterate inexpensively by rapidly synthesizing and testing millions of compounds in parallel, further updating our ML models. This iterative process between predictions and real-world measurements (also known as Active Learning) is similar to the approach DeepMind used to train the AlphaGo model which beat Lee Sedol at Go,” Tilmans explained. “Our lab was built from the ground up to build compounds 10x faster and cheaper than any competing technology while simultaneously improving data quality.”
In announcing the investment round, the company noted that since beginning operations in Fall, 2020, Anagenex has built a special biochemistry lab that can test more than two billion highly drug-like compounds simultaneously using technologies such as DNA Encoded Libraries (DELs) and Affinity Selected Mass Spectrometry. DEL is a technology that the pharmaceutical industry uses to discover small molecules capable of having an effect on biologically relevant targets. Affinity Selected Mass Spectrometry is a method to screen large collections of compounds.
By definition, AI/ML algorithms are designed to answer questions. If you show enough pictures of cows to an AI algorithm, eventually when shown a picture of an animal, it will be able to tell if it’s a cow or not. Anagenex’s AI is similarly designed to answer questions though it’s a more complex question. Here’s Tilmans again, explaining what Anageneex’s AI is predicting and the process thereafter:
“Does the compound bind to a target but not to a closely related target that we do not wish to interfere with. After subsequent experiments, the model is updated to learn which compounds have a biochemical effect. The model seeks to rank compounds in order from best to worst, not just learn a binary “binds” versus “doesn’t bind”.’
Tilmans is only too aware of competitors in the field, small and large. He named a litany of them, a few of whom are Terray Therapeutics, Totus Medicines, 1859 Inc. , Relay Therapeutics, Insitro — and in every case, he felt, unsurpsingly, that Anagenex was superior.
In response to a different question, however, he did acknowledge the prowess of two public companies.
“Very, very few small-molecule drug discovery companies have the combination of large-scale proprietary datasets combined with the computational talent needed to maximally leverage those datasets. Among public companies, Recursion and Exscientia are the only ones. [Chris Gibson, the CEO of Recursion and Richard Law, chief business officer, Exscientia are both taking part in a panel discussion about the state of AI in drug discovery moderated by MedCity News’s senior biopharma reporter Frank Vinluan at our INVEST PharmaTech virtual conference available on-demand on July 26]
Recursion’s clinical pipeline includes clinical-stage drug candidates for neurological diseases cerebral cavernous malformation and neurofibromatosis type 2. The Salt Lake City, Utah-based company is also developing pre-clinical programs for other conditions: Batten disease, Tay-sachs disease and hereditary hemorrhagic telangiectasia.
U.K.-based Exscientia’s platform has resulted in seven drug candidates, three of which are in Phase 1 testing. One of its most advanced internally developed candidate, EXS21546, is an immuno-oncology drug.
Anagenenex meanwhile has “several initial programs focusing on precision oncology and cardiovascular disease,” Tilmans said.
Like so many AI discovery companies leveraging machine learning, the startup is looking for proteins/targets that have been challenging to drug in the past but that have a high disease burden.
Anagenenx is still in the early days of its drug discovery cycle and a drug finding its way to clinic will likely still take a few years, but Tilmans is confident of future success.
“Our platform is a unique synthesis of innovative large scale lab experiments and machine learning,” he said. “Lab experiments alone are slow and expensive but successful in bringing drugs to market. Machine learning is fast and cheap but struggles to bring drugs to market. By combining the strengths of each, we’ve created a unique cost-effective platform to find new medicines.
A current investor was sold on precisely this concept.
“At Lux, we’ve repeatedly seen how merging state of the art computational tools with custom built lab operations transforms drug discovery to bring innovative medicines to patients,” said Zavain Dar, venture partner at Lux Capital. “Still, it’s rare to see companies that blend the two as seamlessly as Anagenex, and we’re thrilled to have been there from the beginning.”
Clearly, Anagenex has investors sold on it vision. Whether it can deliver will only be clear in the long road to commercialization. The first compounds are expected to enter clinical trials phase in two years.
