Oh, the Places You Will Go ... with AI

Had Theodor Geisel (aka Dr. Seuss) still been alive in 2025, he may very well have written a book on the places you can go with AI.

I have the rather vague, poorly described goal of 'learning AI'. At the start of my 6-month reset I knew two things for sure: 1) AI has been and will continue to be built into IT infrastructure changing our lives in crazy ways and 2) I know precious little about the overall AI rapidly unfolding scene, other than NVIDIA was a great buy at $144/share before the stock split 10:1.

So, being so unclear on the landscape of AI, it was difficult to express a solid end point or where exactly to land on a goal. But this week was a great week to advance this.

Over this week, I spend several hours getting cozy with AI, with the help of ChatGPT. I cannot express enough how awesome it was to use ChatGPT to curate this information and walk me through step by step in installing python, PyTorch and other tools. Today, Thursday, I was able to create a 3-D protein structure from a 1-D amino acid sequence. I then was able to test a small molecule drug on the binding affinity of the protein.

In this case, I ....

1) modeled the COVID-19 main protease in 3D using Alphafold

2) downloaded ritonavir in the SDF format from Pubchem

3) Converted it to PDB format using OpenBabel

4) Docked the small molecule into the protein using Swiss Dock.

The binding ranged from -6.8 kcal/mol to -7.1 kcal/mol. It sounds more complicated than it was since ChatGPT was very helpful in referring sites and steps. But this is a first step in doing something I could never do w/o AI, at least not without a biochemistry degree or similar training.

I also spent some time looking at data sets and beginning to write script (using ChatGPT) to model the data in some way. I maybe spent 6-7 hours this week on all this. Really amazing.

To anyone reading this --- there are so many free resources out there virtually anyone can learn this stuff, given a little time and patience. Honestly, the sky is the limit given the free resources out there. This is, again, one of the reasons I wanted to step away: to see what could be done here.

In the end, I want to explore models in drug development (which is where I had worked in some sense for many years) and look at the homelessness issues facing our city and state of NC. I am looking to see what data and models can be built to help serve in this need. I am sure other ideas might pop up, but it is great to head in this direction.

Below are the main sites/resources I used this week:

ChatGPT (love it)

Python

Cuda, cuDNN

PyTorch

Google Colab

Alphafold

Pubchem

Swiss Dock

Open Babel