- Drug Discovery using machine learning is working and holds great promise
- Most cancer treatments are incredibly expensive and only extend life by a few months
- Cancer research started when in WWII it was discovered mustard gas could cause tumor regression
- Lack of diversity in approaches is one of the main bottlenecks
- A lot of friction between a researcher with a great idea and getting funding due to slow, bureaucratic processes
- Underresearched area of chronotherapy, ie timing of drug administration should be further investigated
Sarah Constantin is a mathematician with a PhD from Yale, and worked as a data-scientist for drug discovery at Recursion Pharmaceuticals. Now she is writing a book about cancer research. You can find Sarah and her very informative blog here: https://www.sarah-constantin.org/
Using machine learning & automation to find drugs
Sarah explains how Recursion Pharmaceuticals turns drug discovery into a machine learning & computer vision problem:
Recursion focuses on rare genetic diseases. So they generate thousands of cells that all miss the gene that causes the malfunction in humans suffering from the disease. Then they will screen how thousands of compounds interact with the diseased cells. Through sophisticated computer vision machine learning algorithms that have been trained with datasets of diseased and healthy cells, they can start recognizing healthy cells and thus predict which compounds could turn diseased cells into healthy ones.
Sarah describes Recursion as a tech company whose applying engineering principles to the problem of drug discovery. So they automate a lot of the processes in the lab that would otherwise have to be run manually. So they have high-throughput screening lab, which means that sample preparation is done robotically and can be altered programatically. Sample preparation, staining, drug screening, and microscopy are all automated. Together with turning ‘atoms into bits’ for classification, their cost structure is massively lower than the traditional ways of running assays and they can do it much, much faster.
With this improved process, Recursion can screen thousands of compounds per week very cheaply and effectively. After this, the compound enter the traditional drug discovery process, where these compounds are tested in vivo (animal models) to bring them to Investigational New Drug (IND Status) so they can enter clinical trials.
Is it working? According to Recursion’s pipeline website, the relatively young company has already found a significant number of candidates and are rapidly moving them towards IND status. Investors seem to like what they see and have recently poured $60M into the company in it’s Series B fundraise.
What is the current state of affairs in cancer research?
Cancer Treatment with drugs starts after World War II when it was found out by chance that mustard gas also kills tumors. The National Cancer Institute (NCI) had been formed and they developed the first chemotherapy drug called Mustine (a nitrogen mustard).
Since then, we have been researching cancer and appropriate drugs to cure it for decades, and we have gotten better and preventing it and have made significant progress, but still we’re often still not able to cure cancer.
Targeted therapy has enhanced treatment capability of cancer disease, but still often it doesn’t work – expensive drugs in many cases extend life only for several months. Because cancer is a very complex disease. Not only are there many different kinds of cancer, but even one type of cancer can be very heterogeneous.
So there are a lot of mechanisms within cells to circumvent the therapeutic effect of the drug, for example through angiogenesis – new blood vessels formation to avoid immunosuppressive response.
Read the seminal paper called The Hallmarks of Cancer Sarah mentions.
What is broken in the way we do the research?
#1 Lack of diversity
According to Sarah, one of the main reason we haven’t seen more progress is because there is a lack of diversity in the approaches we are seriously trying out, done by only a handful of large companies. As an example she mentions how many cancer cells use glycolysis instead of cellular respiration to convert sugar into energy. And because this process is less efficient, it means cancer cells use a lot more sugar than regular cells. This means there is a way that drugs can interfere with cells energy supply and so preferentially kill cancer cells. This is a whole class of drugs that is barely being investigated.
Different approaches for cancer treatment exist. A currently very exciting field is that of immunotherapy. There are different approaches in this field, and one important one is based on introduction the patient’s own immune cells which can recognise cancer cells through specific antigens. It is possible to program T cells to destroy cancer, but there is a challenge to physically transfer those cells into particular part of a solid tumor. This is much easier for blood cancers and is where we are seeing outstanding results so far.
#2 Funding availability and friction in getting funding
Bringing a drug from the bench to the beside is extremely expensive and high risk. A recent analysis of 10 cancer drugs through SEC filings put the median number to $648M to get a drug approved. This leads to a sort of an oligopoly of the large pharma companies.
So when researchers come up with new ideas for trials they’d like to do, they have to raise money for that, mostly through writing lengthy grant applications to various governmental and philanthropic funding agencies. Compared to how this was done when everyone was under one roof at the National Cancer Institute, this is very expensive and inefficient. It was very centralized, but a grant application was walking up to someone and say “Hey I want to research this, can I have some money?” and the person could approve it right then and there.
At the time the NCI also always had a layperson involved that would make sure research was strongly focused on getting results, which Sarah believes led to more direct impact of the research on patients compared to what is happening today, were researchers can often be quite removed from doctors and certainly from patients.
What can we do about these problems?
#1 Regulatory Reform
In order to reduce the cost of getting a drug to market, people are working on regulatory reform. There are a number of initiatives focusing on slimming down the FDA’s approval process and create fast-track paths for certain terminally ill people or rare diseases.
#2 New Business & Research Models
Another interesting model is being tried by different groups, including the ChanZuckerberg Intiative, where the pitch to scientists is: “We are business people, we can handle the lawyers, the IP, the negotiations, the marketing etc. so that you can focus on research.” Having those shorter communication channels could dramatically increase the output and impact of those researchers.
Taking drugs at the right time might have a huge influence on efficacy
At bio2040, we always look for unexpected findings and ways to improve the state of affairs. In this case, Sarah brings to our attention very interesting research around a concept called chronotherapy.
According to a group of researchers around Dr. Francis Lévi, timing the drug administration correctly could render the treatment twice as effective and 5 times less toxic versus a suboptimal time. Because metabolism within cells follows a circadian rhythm, cancer cells are less active at a certain time of the day and we can use that property to massively improve the impact of our medicine. We need this low-tech approach to get more attention and get into larger clinical trials.
Read more about Sarah’s work on cancer research here