How can AI understand chemical substances?

Chemicals. Are they safe? How does AI analyze chemicals? There are several AI models that have been developed to characterize chemicals. What do we need to input? The molecular structure. If you provide the molecular structure, the models will find the toxicity of the substance. A molecular structure? In most of case, preparing the molecular structure is a big hurdle. Most of us probably learned it in high school and forgot about it. 
 
The structure of a chemical can drastically change its properties. Toxicity can also change significantly. That's why AI can predict changes in toxicity based on changes in structure. How to prepare molecular structure? Usually, you can draw it., but not in the Paint. There is software specifically designed for the purpose. If you draw the structure in that program, and it converts your molecule into a format that the AI model can use.
 
There are programs available on the web. Below is a database called PubChem. If you access this database, you will see the ability to draw structures.

PubChem first page

 

Click the Draw Structure button and you'll see a window like the one below. This is where you can draw your structure. On the right side, you can draw the structure in the white canvas. Click the various buttons on the left to draw the structure. In the image below, a stick is selected. With this option, you can draw a long chain.

Draw Structure

 

Selecting Hexagon will add a hexagonal ring to the long chain. For practice, I just drew a random structure. Once you've drawn the structure, click the search for this structure button below to find it in the database.

Molecular structure is ready!

 

In this picture, there's a button at the top that says SMILES, and those weird letters are the structural formula. Humans are more comfortable with visual representations, but AI uses information like that. Another representation is called MDL Molfile at the bottom, which is a way to save the structure as a file.
 
Below is a program called ProTox-3.0. Again, you can input a structure and it will predict various toxicity of the input compound.
https://tox.charite.de/protox3/

ProTox-3.0 - Prediction of TOXicity of chemicals

 

ProTox-3.0

 

This home page shows this example structure as a default when you open it. For this structure, various toxicities (hepatotoxicity, neurotoxicity, nephrotoxicity, respiratory toxicity, cardiotoxicity, etc.) will be predicted if you just click the button [Start Tox-Prediction] at the bottom of that page. The program doesn't just search databases. AI model in the website calculates toxicity values for the input structures. We will discuss how it works in more detail later.
 
A molecular structure is usually represented as below. Let's take alcohol as an example. The ethanol in alcohol is what makes you drunk. So, the main effect of the product is due to its active ingredient, ethanol. Ethanol is made up of two Cs (carbon), an O (oxygen), and six Hs (hydrogen), as shown in the image on the top right. Each carbon has a total of four bonds around it. So the first carbon is bonded to three hydrogens and one carbon. The second carbon is connected to two hydrogens, an oxygen, and another carbon. Finally, the oxygen can bond with a total of two neighbors. So it's bonded to one hydrogen and one carbon. However, when a molecular structure is drawn, the hydrogens can be omitted because we can easily guess the number of hydrogens around the carbon. This simplifies the representation of the molecule as shown in the lower right corner of the image below. Since carbon and hydrogen usually make up the majority of the molecular structure, black lines are used to represent the connections between carbons. Where the black line breaks is where the other carbons are located. There is no problem drawing a molecule based on the connections between carbons. You don't have to specify all the hydrogens separately, which saves time.

Ethanol structure

 

 

The active ingredient in Tylenol is acetaminophen. It has the following structural formula. On the right is a representation  with the hydrogen atoms, and on the left is a representation with carbon chains.. They are the same structure. In practice, the left representation is preferred for simplicity, so programs that draw structural formulas also show them in the same format.

Molecular structure of Tylenol