Artificial Intelligence and the (Life) Sciences

AI Life Sciences

Machine learning is becoming an increasingly relevant tool for scientific discovery, including drug discovery and the broader life sciences.

AI has the potential to greatly impact the life sciences, including drug discovery. It can help analyze large amounts of biological data and make predictions that improve scientific discovery.

One of the main obstacles is the complexity of biological systems. Unlike physical systems, which can be accurately modeled and predicted using AI, biological systems are much more difficult to comprehend and simulate. This complexity makes it difficult for AI algorithms to accurately predict the behavior of biological systems, which is crucial for the development of new drugs.

Another challenge is the limited amount of data available for training AI algorithms. In order to make accurate predictions, AI algorithms require large amounts of high-quality data. However, the life sciences field has historically been slow to generate and share data, which makes it difficult for AI algorithms to learn from it.

Potential solution include better moleculear representations and tools to get higher resolution data. In addition, researchers are working to improve the availability and quality of biological data through initiatives such as the Open Science movement, where we need to generate more and better data available to everyone.

A Future History of Biomedical Progress

So where are we with deep learning for biochem? by Lada Nuzhna

“Despite its potential, the use of Machine Learning in drug discovery is in its early stages. Current molecular representations don’t contain enough relevant information to fully specify chemical structures, and molecular tools are still bottlenecking the resolution of things. The accuracy of drug properties is limited by the resolution of molecular tools, biases from sample processing, and the complexity of biological systems.

There are limitations of using deep learning in biochemistry and some potential reasons for its slow progress. Among the main problems are:

representation of molecular data, neither of the popular methods contains enough relevant information to fully specify chemical structure
complexity of biological systems, molecular tools still bottleneck the resolution of things

Additionally, there is a challenge in combining multi-omics data due to disparities in timescales, the absence of measurements, and the heterogeneous nature of individual cells. Protocols are being created to measure multiple modalities from single cells, but further research is necessary to comprehend the mechanisms involved.

Moreover, there needs to be a shift from simply identifying correlations to gaining data that answers significant queries. I am pessimistic here, but I still recognize the positive aspects of the situation (my outlook is best expressed by Derek Lowe’s “short-term pessimist, long-term optimist” attitude). I believe that DL/ML should be employed to supplement experiments, not replace them. My other wish is for the field to stop focusing solely on issues that ML does well in biology (such as discovering more correlations - everything in a cell is correlated!) and begin focusing on”

A Future History of Biomedical Progress by Adam Green

progress in basic biology research tools has created the potential for accelerating medical progress; however, this potential will not be realized unless we fundamentally rethink our approach to biomedical research. Doing so will require discarding the reductionist, human-legibility-centric research ethos underlying current biomedical research, which has generated the remarkable basic biology progress we have seen, in favor of a purely control-centric ethos based on machine learning. Equipped with this new research ethos, we will realize that control of biological systems obeys empirical scaling laws and is bottlenecked by biocompute. These insights point the way toward accelerating biomedical progress. […]
One cut on this is how physics-like you think the future of biomedical research is: are there human-comprehensible “general laws” of biomedical dynamics left to discover, or have all the important ones already been discovered? And how lumpy is the distribution of returns to these ideas—will we get another theory on par with Darwin’s?
For instance, RNA polymerases were discovered over 50 years ago, and a tremendous amount of basic biology knowledge has followed from this discovery—had we never discovered them, our knowledge of transcriptional regulation, and therefore biomedical dynamics, would be correspondingly impoverished. Yet when was the last time we made a similarly momentous discovery in basic biology? Might biomedicine be exhausted of grand unifying theories, left only with factoids to discover? Or might these theories and laws be inexpressible in the language of human-legible models?

A lot of relevant ml bio is about predictions (eg. AlphaFold), and intricately connected with each other

predicting the

structure of a protein based on its amino acid sequence.
function of a protein
folding of a protein.
3D structure of a protein.
interaction between proteins.
expression of a gene.
function of a gene
sequence of a gene.

—

Videos, Courses

AI is Industrializing Discovery by Vijay Pande

Using AI to accelerate scientific discovery - Demis Hassabis (Crick Insight Lecture Series)

AlpfaFold

MIT CompBio Lecture Series

Daphne Koller (Insitro, Calico) on ML Drug Discovery

Bharath Ramsundar (DeepChem) — Deep Learning for Molecules and Medicine Discovery

—

These books are great starters:

Vincent Weisser