Acceleration

I keep my finger on the pulse of progress. Lately, I’ve seen a groundswell of articles with a common theme: Deep learning is disrupting everything and it is accelerating. Here are some examples from just the last month:

• AlphaFold solves protein folding
• Caltech’s Covid AI outperforms other models
• Nvidia doubles RAM in A100 to 80gb
• Cerebras CS-1 outperforms supercomputer

All of this against the backdrop of AI regularly outperforming medical experts:

• CNN outperforms 58 dermatologists on cancern diagnosis
• AI outperforms six radiologists on breast cancer x-rays

The hits just keep coming faster and faster.

Medical Implications

Genetic Screening on Steroids

Technologies like AlphaFold mean that we will soon be able to simulate the folding of every genetic variant in your genome. Genes are long strings of DNA with polymorphisms - letters that have been swapped. We can read these markers now with genetic sequencing but we can only use statistics to try and identify what the ramifications are. Services like 23andMe and Genomelink use crowdsourced research and statistical comparison to say what you are likely to have. While valuable, this methodology leaves a lot to be desired. It does not answer important questions like “Why?” - Why does one gene variant result in disease while another does not? In the past, every gene variant would require tremendous amount of investigation. Researchers would need to spend months or years in the lab to ascertain the ground truth; how does one gene variant cause proteins to misfold?

Protein folding is like nanomachine origami. If the mitochondria is the powerhouse of the cell, proteins are the building material and the tools that build the cell. Due to the size and complexity of proteins, there are as many as several billion different kinds. Investigating each one individually is an intractable problem. That becomes less true with AlphaFold. Furthermore, new processing technology like Nvidia’s A100 and Cerebras CS-1 lower the barrier. AlphaFold-as-a-Service very well could be a thing within the next few months. Send in your DNA profile and it will infer all the protein misfolds in your entire genome.

Drug Candidates Galor

The cost of new drugs goes up exponentially with time. That’s a problem. Technologies like AlphaFold won’t replace the entire drug pipeline, but it will allow drug companies to survey tens of thousands of candidates before every getting into the lab. Beyond just identifying drug candidates way ahead of time, I anticipate that future iterations of AlphaFold will be able to simulate the interactions of multiple proteins. The one-two punch of genetic sequencing and mastery of protein folding should result in the birth of truly personalized medicine.

Chronic Disease is History

Age-related diseases are almost all caused by misfolding proteins. Alzheimer’s, for example, is caused by plaques that accumulate due to an inability to clear them out. The same is true of most cardiovascular disease - hence why some people are impervious while others suffer heart attacks and stroke before their fiftieth birthday. Cancer, one of the scariest diseases of all, is no different. Fairly soon, I suspect we will learn to utilize the genetic machinery throughout our bodies to cure all age-related disease and possibly aging itself. None of these problems can be tackled by a single drug as they are multifaceted problems with complex interactions. Personalized medicine will be mandatory in order to even make use of these technologies.

Healthcare Costs Under Control

An ounce of prevention is worth a pound of cure.

But in the arena of modern medicine, it’s usually orders of magnitude cheaper. Degenerative and age-related diseases are expensive. The most expensive diseases are cardiovascular disease, cancer, and metabolic diseases such as diabetes and obesity. Imagine a world where a simple blood test allows the pharmacy to create a custom prescription for you and you know for a fact that you never have to worry about these diseases. Imagine - $10 a month for a customized pill that guarantees you’ll never get dementia, stroke, diabetes, or cancer.

The best crisis is the one that never happens. We are about to live through the biggest medical revolution since antibiotics and vaccines combined.

Broader Implications

The recent news proves that AI and Deep Learning are disruptive in major ways. These technologies are no longer limited to iterative improvements on image recognition and other simpler problems. We’re now seeing the needle move in big ways. It’s only a matter of time before Deep Learning is fully cemented in as the next generation of numerous technologies.