Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Molecular Identification via Molecular Fingerprint extraction from Atomic Force Microscopy images

Manuel González Lastre, Pablo Pou, Miguel Wiche, Daniel Ebeling, Andre Schirmeisen, Rubén Pérez

TL;DR

This work addresses molecular identification from HR-AFM images by predicting Extended Connectivity Fingerprint (ECFP4) descriptors from 3D HR-AFM stacks and performing virtual screening against a molecular database using Tanimoto similarity. To overcome information loss from hashing, it combines a secondary DL model that predicts the chemical formula, boosting identification accuracy to 97.6%, and demonstrates the method on a large simulated QUAM-AFM dataset. The approach enables a confidence-ranked list of candidate molecules and shows promising transfer to experimental images, with attention to substrate-induced corrugation effects. The work advances molecular identification in on-surface chemistry by linking high-resolution AFM imaging to a robust, substructure-based descriptor.

Abstract

Non--Contact Atomic Force Microscopy with CO--functionalized metal tips (referred to as HR-AFM) provides access to the internal structure of individual molecules adsorbed on a surface with totally unprecedented resolution. Previous works have shown that deep learning (DL) models can retrieve the chemical and structural information encoded in a 3D stack of constant-height HR--AFM images, leading to molecular identification. In this work, we overcome their limitations by using a well-established description of the molecular structure in terms of topological fingerprints, the 1024--bit Extended Connectivity Chemical Fingerprints of radius 2 (ECFP4), that were developed for substructure and similarity searching. ECFPs provide local structural information of the molecule, each bit correlating with a particular substructure within the molecule. Our DL model is able to extract this optimized structural descriptor from the 3D HR--AFM stacks and use it, through virtual screening, to identify molecules from their predicted ECFP4 with a retrieval accuracy on theoretical images of 95.4\%. Furthermore, this approach, unlike previous DL models, assigns a confidence score, the Tanimoto similarity, to each of the candidate molecules, thus providing information on the reliability of the identification. By construction, the number of times a certain substructure is present in the molecule is lost during the hashing process, necessary to make them useful for machine learning applications. We show that it is possible to complement the fingerprint-based virtual screening with global information provided by another DL model that predicts from the same HR--AFM stacks the chemical formula, boosting the identification accuracy up to a 97.6\%. Finally, we perform a limited test with experimental images, obtaining promising results towards the application of this pipeline under real conditions

Molecular Identification via Molecular Fingerprint extraction from Atomic Force Microscopy images

TL;DR

This work addresses molecular identification from HR-AFM images by predicting Extended Connectivity Fingerprint (ECFP4) descriptors from 3D HR-AFM stacks and performing virtual screening against a molecular database using Tanimoto similarity. To overcome information loss from hashing, it combines a secondary DL model that predicts the chemical formula, boosting identification accuracy to 97.6%, and demonstrates the method on a large simulated QUAM-AFM dataset. The approach enables a confidence-ranked list of candidate molecules and shows promising transfer to experimental images, with attention to substrate-induced corrugation effects. The work advances molecular identification in on-surface chemistry by linking high-resolution AFM imaging to a robust, substructure-based descriptor.

Abstract

Non--Contact Atomic Force Microscopy with CO--functionalized metal tips (referred to as HR-AFM) provides access to the internal structure of individual molecules adsorbed on a surface with totally unprecedented resolution. Previous works have shown that deep learning (DL) models can retrieve the chemical and structural information encoded in a 3D stack of constant-height HR--AFM images, leading to molecular identification. In this work, we overcome their limitations by using a well-established description of the molecular structure in terms of topological fingerprints, the 1024--bit Extended Connectivity Chemical Fingerprints of radius 2 (ECFP4), that were developed for substructure and similarity searching. ECFPs provide local structural information of the molecule, each bit correlating with a particular substructure within the molecule. Our DL model is able to extract this optimized structural descriptor from the 3D HR--AFM stacks and use it, through virtual screening, to identify molecules from their predicted ECFP4 with a retrieval accuracy on theoretical images of 95.4\%. Furthermore, this approach, unlike previous DL models, assigns a confidence score, the Tanimoto similarity, to each of the candidate molecules, thus providing information on the reliability of the identification. By construction, the number of times a certain substructure is present in the molecule is lost during the hashing process, necessary to make them useful for machine learning applications. We show that it is possible to complement the fingerprint-based virtual screening with global information provided by another DL model that predicts from the same HR--AFM stacks the chemical formula, boosting the identification accuracy up to a 97.6\%. Finally, we perform a limited test with experimental images, obtaining promising results towards the application of this pipeline under real conditions
Paper Structure (13 sections, 2 equations, 9 figures)

This paper contains 13 sections, 2 equations, 9 figures.

Table of Contents

  1. Introduction
  2. Methods
  3. Molecular fingerprints
  4. Tanimoto similarity and virtual screening
  5. Architecture of models
  6. Training and evaluation
  7. DFT calculations and simulation of HR--AFM images
  8. Results and discussion
  9. Predicting ECFP4 from HR--AFM images.
  10. Molecular Identification via virtual screening
  11. Effect of the adsorption-induced molecular corrugation
  12. Experimental images
  13. Conclusions

Figures (9)

  • Figure 1: Diagram of the molecular identification pipeline. From the experiment, we obtain the 3D HR--AFM stack consisting of 10 constant--height images, which is fed to our neural network to extract the Extended Connectivity Topological Fingerprints (ECFP4). Then, we perform a virtual screening with the predicted fingerprints against a molecular database molecule/fingerprints pairs and rank by decreasing tanimoto similarity.
  • Figure 2: Tanimoto similarity distributions between predicted and ground truth fingerprints (red) and randomly drawn molecules fingerprints (blue). We can extract enough chemical information to distinguish a molecule from the bulk.
  • Figure 3: Identification accuracy versus corrugation. We compute the accuracy for molecules with corrugation $<$ 25 pm (green), 25--75 pm (orange) 75--125 pm (purple) and $>$ 125 pm (magenta). Dashed black lines represent the accuracy over all corrugation groups. Enriching the virtual screening with the chemical formula improves accuracy across all corrugation groups.
  • Figure 4: Examples of identification of polycyclic aromatic hydrocarbons over theoretical 3D stacks. Columns from left to right, constant-height AFM images (1-3), ground truth molecule (4) and top (5) and second (6) candidates. Under each candidate, tanimoto similarity, $S$ and corrugation, $\Delta z$ is expressed. Molecules from first to last row are Tetrabenzo(a,c,g,s)heptaphene (CID:143932), Benzo[1,2,3-bc:4,5,6-b'c']dicoronene (CID:636081) and Tetramethyl-Undecacyclo-Tetraconta-Icosaene (CID: 59721948), where methyl groups have been highlighted. The model identifies the correct molecules with high confidence.
  • Figure 5: Examples of identification of molecules with nitrogen, oxygen and sulfur atoms. From first to last row: 4,4'-Bi[1,2,3-thiadiazole] (CID:2748722), 5-methyl-2-(2H-triazole-4-carbonylamino)thiophene-3-carboxylic acid (CID:63616469) and 5-Pyrazolo[1,5-a]pyridin-3-yl-1,2,4-oxadiazole-3-carboxylic acid (CID:103122053). IN the last two rows, the differences between candidate molecules have been highlighted to guide the reader.
  • ...and 4 more figures