Computer modeling of Molecules

A new collection of artificial intelligence algorithms can allow software to rapidly determine a molecule’s properties based on the theoretical structure of the molecules. With this tool chemists could test molecules out on computers avoiding finding them by way of trial-and-error experiments.

The big problem for computer-aided design chemistry is Schrödinger’s equation. While it could theoretically be solved find all of a molecule’s chemical and physical properties the complexity of the equation increases with the number of protons and electrons in the atom or molecule. As a result there are only exact solutions for simple systems such as hydrogen atoms and hydrogen molecules. The level of complexity makes it impossible to precisely predict properties of large molecules need in medicine and engineering.

The solution was to skip Schrödinger’s equation and go with machine learning algorithms that are already being used to find large data set patterns that are controlled complex rules This new collection of algorithms will use existing molecules as examples and use the patterns to predict the properties of new molecules

The team created a 7165 molecule database and then used 1000 of them for the computer to find structural patterns that could predict chemical properties. They then used the other 6165 molecules to test the resulting algorithm and they produced values with in 1% of the known value which is comparable to earlier methods. However the new algorithm did it in a millisecond and not the hour other methods would take.



Sponsor a page

at $1 a month

$11 for a year

Support this website  with a $1 gift.


Visit our

Online Store

Gifts of other amounts

Click Here


Custom Search