13
$\begingroup$

In computational chemistry, I have come across many test sets which are used to benchmark a new method. For example, the G2 test set for thermochemical calculations (heats of formation). There is the SAMPL-1 set for hydration free energies. These sets usually have a diverse range of molecules/species of interest so that it is possible to check if the new method has a problem in modelling a particular class of molecules.

I am currently working on predicting UV-vis absoprtion band of organic molecules (i.e. the $\lambda_{max}$) with statistical learning. So I was wondering if there is a robust benchmark test set with UV-vis data (with experimental results) which I can use for my purpose. I haven't found much with google searches so far.

$\endgroup$
1
  • 4
    $\begingroup$ Hopefully an answer will address possibilities/limitations to account for solvatochromism (batho/hypsochromic shift), too. (+1) $\endgroup$
    – Buttonwood
    Commented Oct 18, 2021 at 11:35

1 Answer 1

12
$\begingroup$

There is a program called PhotochemCAD by Tanuchi and Lindsey which contains a database of experimental absorption and emission spectra of more than 300 organic and organometallic compounds. The spectra are available in plain text form. While the maximum absorption wavelengths of the most intense peaks of each compound have already been tabulated in the main text of the linked article, if you are also interested in the absorption maxima of weaker peaks too, you may need to extract this information from the spectra by yourself.

By the way, given that the database contains complete spectra information, it may actually be a good idea to train your model to reproduce the overall line shape instead of just the absorption maxima, giving an even more useful model :)

Taniguchi, M.; Lindsey J. S. Database of Absorption and Fluorescence Spectra of >300 Common Compounds for use in PhotochemCAD in Photochem. Photobiol. 2018, 94, 90-327; doi 10.1111/php.12860.

Project page: https://www.photochemcad.com/

$\endgroup$
2
  • $\begingroup$ Thanks, that is very helpful. I am not sure how I can train a model to reproduce the whole lineshape though, so any ideas/leads in that direction would be also useful. (I am currently setting the lambda_max as the target of the machine learning algorithm) $\endgroup$
    – S R Maiti
    Commented Oct 20, 2021 at 8:52
  • 1
    $\begingroup$ @SRMaiti For example you can make the output layer of the machine learning model generate predictions of extinction coefficients of each wavelength, and fit the model against the PhotochemCAD data $\endgroup$
    – wzkchem5
    Commented Oct 20, 2021 at 17:07

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .