AFM Molecular Identification Using Neural IUPAC Name Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to accurately identify molecules through atomic force microscopy (AFM) images, particularly for non-planar structures, and existing AI techniques fail to generalize beyond training data, limiting molecular identification to known molecules.
Innovation Solution
A method using two trained Multimodal Recurrent Neural Networks (M-RNN and AM-RNN) to generate IUPAC names from AFM images, comprising convolutional and recurrent neural networks, with pretraining and data augmentation to enhance generalization and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning classification is used to identify molecules from AFM images, then classification accuracy for known molecules is improved, but the method fails to generalize to molecules outside the training dataset
Solution Approach 1:
The patent replaces traditional image classification approaches with a language model-based system that translates AFM image features into molecular structure descriptions (SMILES notation). This substitution enables the system to generate names for any molecule based on its structural features rather than matching against a predefined classification database, thereby achieving both high accuracy and broad generalization capability.
Solution Approach 2:
The language model is trained to perform multiple functions: identifying molecular structures, generating SMILES notation, and producing IUPAC names. This multi-functional approach allows a single system to handle diverse molecular types and structures, making the identification method universally applicable to any molecule regardless of whether it exists in the training data.
2Ease of operation
If visual inspection by human eyes is used for molecular identification, then interpretability is maintained, but the task becomes impossible due to the huge variety of chemical environments
Solution Approach 1:
The patent introduces a language model as an intermediary between AFM image analysis and molecular identification. The model translates complex image features into human-readable molecular structure descriptions and names, maintaining interpretability while handling the enormous complexity of molecular recognition that would be impossible for human visual inspection.
3Adaptability or versatility
If a large dataset is used for training to cover all possible molecules, then classification coverage is improved, but computational challenges become intractable
Solution Approach 1:
Instead of training on a massive dataset of all possible molecules and then classifying, the patent inverts the approach by training a language model to generate molecular names from structural features. This allows the system to handle any molecule theoretically possible, eliminating the need for exhaustive training data while keeping computational requirements manageable.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate identification and naming of molecules from AFM images, overcoming limitations of existing AI methods by providing a systematic and general approach to molecular recognition.
Implementation Method 1
The outstanding contrast arises from the Pauli repulsion between the CO probe and the sample molecule
Implementation Method 2
modified by the electrostatic (ES) interaction between the potential created by the sample and the charge distribution associated with the oxygen lone pair at the probe
Implementation Method 3
Atomic Force Microscopy (AFM) operated in its frequency modulation (FM) mode allows the characterization and manipulation of all kind of materials at the atomic scale. This is achieved measuring the change in the frequency of an oscillating tip due to its interaction with the sample
Data Source
Figure 1A~1H
Figure 2
Figure 3
AI summary
The invention relates to a computer implemented method for identifying a molecule from Atomic Force Microscopy images and generating the name of the molecule according to the IUPAC nomenclature using two trained Multimodal Recurrent Neural Networks. Furthermore, the present invention refers to the system configured to carry out the steps of said method. The present invention is therefore of interest in the area of nanotechnology, particularly in areas related to on-surface chemical reactions and therefore of interest for the Atomic Force Microscopy users and manufacturers.