Christopher Trummer, co-founder and CEO of tech-enabled early-stage drug discovery company CelerisTX. He shares his insights with Health Tech World about how artificial intelligence will “disrupt” how we treat Parkinson’s Disease…
Artificial intelligence (AI) and machine learning (ML) are already having a major impact on the biopharma space. They are being used to change the ways in which we develop new therapeutics, identify biomarkers, and treat patients. And at this point, the extent of their potential can only be guessed at.
But for diseases that are currently considered incurable, ML and AI have a potentially significant role to play, helping to identify the hidden links between the disease and biological causes, while deriving therapies that will potentially lead to the production of curatives.
“Transformative” for Parkinson’s
And for Parkinson’s and other central nervous system (CNS) diseases, the impact could be transformative.
Pharmaceutical therapeutics & AI
It’s no secret that the path towards pharmaceutical development can be complex and convoluted. To simplify, it can be broken down into four major stages.
- Biological target identification – such as a protein within the body that causes a disease – and validation
- Compound (molecule) screening and lead discovery
- Preclinical development
- Clinical development
While the latter stages have already become known to the broader public, especially through the COVID-19 crisis, the first two stages are also co-decisive. Machine learning can now help with all four stages.
Using a wide variety of algorithms, data sources and application areas, it can support both predictive models and generative models to develop new drugs.
How can this translate to Parkinson’s treatment?
Through new predictive models and exploiting data sources, researchers are able to make hidden connections between a disease to a target. This gives us more potential targets to eliminate incurable diseases from multiple levels.
These target-disease associations, or gene-disease associations, can only be obtained by exploiting high-dimensional omics data.
What is omics data?
These are laboratory-derived data on DNA, RNA, protein, and metabolites, and are commonly referred to as genomics, transcriptomics, proteomics and metabolomics. As a result, in addition to the algorithms of machine learning models, laboratory technology is improving exponentially, giving researchers access to high-resolution, error-free data at a high throughput.
And it is these data sources that are critical to generating new biological insights and thereby identifying targets.
ML is the only way to digest high-dimensional data in a high throughput, make deductions from it, and identify patterns. With millions and billions of data points involved, it’s a process that is beyond the abilities of a human.
For phase 2 of therapeutic development, compound design, several approaches are now being pursued to cure currently incurable diseases.
One approach is that of ours at Celeris Therapeutics, which is developing a holistic platform to create new chemical matter that goes beyond the traditional therapeutic modality of inhibition.
Of particular note, is the degrader approach, in which instead of inhibiting certain metabolic processes, one seeks to directly degrade parts of the metabolic process chain.
This field is called Targeted Protein Degradation and makes use of already natural human processes (protein biosynthesis and protein degradation).
Making use of machine learning
One prejudice against Machine Learning is the necessary use of large quantities of data (as used in target identification). This is another area that we are challenging, with the application of Active Learning.
Therapeutic development at the compound design stage is again broken down into Design-Make-Test-Analyse phases.
We are leveraging its models to design new compounds, for synthesising, testing, and sending those results back into design in a closed loop.
Through this iterative approach, new solutions are proposed in a targeted manner, while new data is generated. Due to the focus on a single problem, in such a procedure, smaller amounts of data are needed. \
And from a large number of degrees of freedom for a design of new compounds, an ever-decreasing set of chemical space is evaluated.
Thus, one goes from a broad question to a deep question within a few iterations, whereby fewer data points can already make the difference. And that’s where the treatment of Parkinson’s comes in.
How exactly can degraders cure Parkinson’s?
First, we need to look again at the biology of Parkinson’s disease. Parkinson’s patients, like Alzheimer’s patients, suffer from fibril formation of some proteins. In Alzheimer’s, it is the famous tau protein, in Parkinson’s it is alpha-synuclein. Alpha-synuclein is therefore in direct target-disease association.
Now, with alpha-synuclein, the situation is complicated for normal and conventional drugs because they work via inactivation by inhibition.
However, this does not work here, because the fibrils are degenerated structures, which also no longer perform any actual metabolic function. They are simply fibrils that accumulate and lead to irreversible damage.
Degraders can now do one thing: they allow these fibrils to be virtually eaten up, which no other modality may accomplish. And that’s going to be the game-changer against Parkinson’s.
AI & ML in the medical sphere
There is always reluctance when you attempt to introduce AI and ML into the medical sphere.
By AI and ML applications can give developers and trial teams the support they need to make more accurate decisions, streamline the development process, and address previously untreatable conditions. And that has to be embraced for the good of all.
Christopher Trummer
Original post: https://www.htworld.co.uk/news/how-ai-will-disrupt-the-treatment-of-parkinsons/