Machine learning modeling of electronic spectra and thermodynamic stability for a comprehensive chemical space of melanin

Abstract

Melanin as a bio-optoelectronic material holds immense potential. However, understanding its exact molecular structure has been stalling for decades due to difficulties in experiments, which hinder uncovering its structure-property relationship. Conventional theoretical modeling is also limited due to the huge size of its chemical space, which results in millions of possible oligomer structures. Here, we design a comprehensive virtual chemical space of melanin oligomers and develop a machine learning-based approach for predicting their entire UV-visible spectra and thermodynamic stability using fingerprint input. We also show the similarity of our predicted Boltzmann-weighted average spectrum with the experimental spectrum and discuss their potentiality towards bio-optoelectronics.