CLAMP: A Contrastive Language and Molecule Pre-Training Network for Scaling Artificial Photosynthesis Candidate Identification
Overview
Artificial photosynthesis is the process of turning sunlight and CO2 into chemical products. This is especially desirable due to the ability for carbon neutral fuel and scalable carbon dioxide reduction. Metal Organic Frameworks (MOFs) are porous crystals that can be synthesized to have various active properties—carbon capture, water treatment, or even artificial photosynthesis. The largest obstacle to current research and development of crystals is finding new sites & reactions to synthesize crystals. Machine learning is applied in this space due to its ability to find new permutations that can be tested. For artificial photosynthesis there are no candidates for visible light artificial photosynthesis, which makes machine learning with prior data almost impossible. Similar problems are faced in image generation where due to low performance small amounts of specific data, large amounts of unsupervised generic text and image data are used to generate images with high accuracy. Therefore the engineering goal is to build an unsupervised contrastive language and molecule pre-training framework which could predict possible MOF photo-catalysts. Building a web scraper that gathered 222k crystal text pairs from research papers, an unsupervised zero shot classification model was trained taking advantage of linguistic structure. Without any specific training data, an ~82% accuracy was achieved and ~75% accuracy for photocatalyst prediction. With no prior specific data, this could be a breakthrough technology. This novel network can be cross applied to any task that can be described via text, opening completely new methods to think about chemical generation.
Competition history
- ISEF 2023
Resources
Related projects
ISEF · 2022
CarbNN: A Novel Active Transfer Learning Neural Network to Build De Novo Metal Organic Frameworks (MOFs) for Carbon Capture
ISEF · 2021
A Perovskite Crystal Structure Prediction and Screening System Using Complex Machine Learning Methods
ISEF · 2022
Creation of a Predictive MOF Performance Indication System Utilizing Synthesis Criteria and Machine Learning
ISEF · 2025
De novo Protein Generation for Enhanced Efficiency in Biohybrid Solar Cells
ISEF · 2026
Machine Learning Guided and Accelerated Development of Semiconductor Photothermal Materials Using Doping Strategy to Modify Bandgap for Enhancing Solar Water Evaporation
ISEF · 2019
Discovery of FAZnF3, a Hybrid Organic-inorganic Perovskite for Photocatalytic Water Splitting
ISEF · 2026
SolarEye: Novel Solar Energy Harvesting Retinal Molecule Derivative and Rhodopsin Inspired Protein Scaffold Biomaterial Designed by Utilizing a Large-Language-Model and Computational Tools
ISEF · 2023
A Novel Adversarial Autoencoder With Latent-Based Optimization to Design De-Novo Ionic Liquids for Post-Combustion Carbon Capture
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: Regeneron International Science and Engineering Fair