Intelligent Molecules: Applications of Machine Learning in Chemistry

Machine learning is an aspect of computing that falls under artificial intelligence that allows information to be collected, digested, and used in the future. There are two types of machine learning: supervised and unsupervised, which differ in if the answer that we will get is known or not already.  Machine learning is used primarily in computers and online systems to save preferences and tailor the experience to the human user. In those application, the “machine” acts as if it has a brain and remembers information so it can regurgitate it later in a situation in which it would make sense, as people do in conversations.

However, not all applications of artificial intelligence must be in computers or online systems. Computer scientists and biological engineers at Harvard have been trying to apply machine learning to molecules that could eventually “that can automatically detect, diagnose, and treat a variety of diseases using a cocktail of chemicals.”

The mathematics and biology necessary to input machine learning into molecules.

Nanotechnology has become a larger focus more recently, so it would be reasonable for the next step in machine learning to move from the macroscopic to the microscopic level, and to biological systems. Currently,  therapeutics do not change in response to the body, they only carry out their predetermined task. If drugs could respond to the body, human error in diagnosis and treatments would be reduced. In research, molecules would be able to relay better data would be able to be collected and characterized better than chemical reactions can be now, since it is difficult to study reactions in situ.

In the future, I believe machine learning implemented in molecules could revolutionize therapeutics for mental illnesses such as bipolar disorder, which changes more frequently than other mood disorders such as depression. Researchers could always use extra help, why not have the molecules and biological systems they are studying do just that?

More biologically and chemically relevant applications of technology can be found at the Wyss Institute which contributed biological engineers to the project of implementing machine learning into molecules.

Sources:

Article: Perry, C. (2013). Programming smart molecules. Harvard.

Picture: Napp, N., Adams, R. P. (2013). Message Passing Inference Networks with Chemical Reaction Networks. NIPS Proceedings, 1-9. https://www.seas.harvard.edu/news/2013/12/programming-smart-molecules

Interview with a Professional Development Chemist

I was lucky enough to interview a development chemist this week. Jordan Wilkins* focuses on the DOE - design of experiment - to provide the exact colors of paints that their clients want, for industrial purposes. However, providing the correct color in a reliable and lasting paint is not a simple task. Paints are based on resins, which are organic molecules that come from plants and other natural materials and often need to be switched out to fit specific needs. Other compounds are also used to achieve the wanted results such as silica, iron (III) oxide - which is used for pigment - and xylene - which is used as a solvent for some paints. It is likely that you have experienced xylene before in your everyday life, as it is what is used in some permanent markers and makes their distinct smell.

iron (III) oxide (1)

silica (2)

xylene

different isomers of xylene

A balance of hydroxides and hydrogen ions must be kept to maintain the proper pH for the paint to work. The paints must also be suited to the conditions in which they will be used. As paint dries, a reaction is occurring aided by catalysts, in which the reaction must complete at the same pace as the catalyst. 

It can be a long and complicated process to figure out what reactions must occur to develop the wanted product, but technology can help developers like Wilkins with the process. One of the main pieces of technology used in paint development are computer programs that, when given an input of a desired product based on color and conditions, will give an output of many reaction options that will likely work to reach that product. These computer programs are known as formulation tools, and also assist in figuring out how to balance hydroxide and hydrogen ions, in narrowing down which of the output reactions are more realistic for the purpose given, which resins to switch to, as well the reaction rates present and needed. Every time a change is made to the reaction process, that change can be input into the programs so a sample tree can be developed and the programs will determine if the change is substantial. An example of this would be if the temperature of a reaction is changed, sometimes that change is not drastic enough to actually modify the reaction because most reactions can happen under a range of temperatures, not just one singular temperature. Once Wilkins knows what reactions might work, they can run large batches of reactions and continue to give the computer programs more input so they can continue to help Wilkins narrow down their options.

The formulation tools involve the use of on-site databases that are private to their company, which currently are based on spreadsheets. Sometimes, Wilkins and their fellow research chemists must go in and edit the spreadsheets to get the results they need, and as a way to upgrade their system. In the future, the databases will be moving into web-based systems by another department at Wilkins' company so they will not have to use code to develop it. However, Wilkins and other employees at the company will be able to use the databases on any company-approved device through that device's web browser, which will include mobile technologies. The main benefit of having mobile access to the databases is that the employees that visit customers and receive specific questions about their paint can use those databases on-site instead of having to call the employees that are back on computers within the company's system. The only outside databases that Wilkins' company uses are industry standards for necessary information.

Once Wilkins has developed a new paint sample, a spectrophotometer can be used to test that the color the customer has asked for was made. Spectrophotometers are a very standard piece of technology used by chemists at all levels, as they vary in complexity and applications. Spectrophotometers generally work by sending light to a sample to test absorption of certain wavelengths. For industrial purposes, a spectrophotometer that is able to read metallic colors must be used, since industrial paints will dry to produce films that have both shine and pigment qualities that will change under different angles and light effects on their metal surfaces.

example spectrophotometer for metallic colors (3)

how metallic paints will react with light to produce

their color and shine once it has dried to its surface (4)

Though spectrophotometers can provide readings on screen, the data usually needs to be furthered analyzed using a software program on a separate computer system. An example software program is developed by Cyberchrome, whose's products can be found at their website: http://www.cyberchromeusa.com/products.



*Name changed to protect privacy. The interview responses reflect only the opinion of the interviewee, and not the formal position of their unnamed company or the industry they work in.

Chemical structures were made with ChemDraw.

References:
1. https://en.wikipedia.org/wiki/Iron%28III%29_oxide#/media/File:Iron%28III%29-oxide-sample.jpg
2. http://www.chm.bris.ac.uk/motm/silica/silicah.htm
3. https://www.byk.com/en/instruments/products/?a=2&b=18&f=0&faction=#fam1
4. http://blog.xrite.com/using-spectros-for-print-packaging-manufacturing/