Molecular identification with atomic force microscopy and conditional generative adversarial networks

High-Resolution Atomic Force Microscopy (HR-AFM) using CO-functionalized metal tips offers unprecedented resolution for imaging molecular internal structures. This paper proposes a Conditional Generative Adversarial Network (CGAN) to extract chemical information from HR-AFM images for complete molecular identification. The CGAN converts a stack of constant-height HR-AFM images at various tip-sample distances into ball-and-stick depictions, representing chemical species with balls of different color and size and bonds with sticks. Trained and tested using the QUAM-AFM dataset (686,000 organic molecules), the CGAN demonstrates accuracy in identifying molecules from both theoretical and experimental AFM images.

Jaime Carracedo-Cosmé, Rubén Pérez