HUGE breakthrough! This AI discovers unknown molecules

Key Concepts

• Molecules: Fundamental building blocks of all matter, including living organisms.
• Natural Molecules: Molecules found in nature, many of which are yet to be identified.
• Tandem Mass Spectrometry (LCMS): A technique used to identify molecules by separating and fragmenting them, creating a unique spectrum or "molecular fingerprint."
• Spectra: Visual representation of the fragmented molecules produced by mass spectrometry.
• Self-Supervised Learning: A machine learning technique where the AI learns from unlabeled data.
• Dreams Neural Network: An AI model developed to interpret molecular spectra and map them to molecular properties.
• Dreams Atlas: A multi-dimensional map representing the relationships between 201 million natural molecules based on their spectra and predicted properties.
• GNPS (Global Natural Products Social Molecular Networking): A database containing millions of molecular spectra.
• Molecular Embedding: The position of a molecule within the Dreams Atlas, reflecting its similarity to other molecules.
• Lipinski's Rule of Five: A set of guidelines to determine if a molecule is likely to be an orally active drug.
• Plant Metabolite: A chemical compound produced by a plant.

1. The Unexplored Chemical Universe

• Life is composed of molecules, but less than 10% of natural molecules have been identified.
• Identifying the remaining 90% could lead to breakthroughs in disease diagnosis, drug discovery, longevity, new chemicals for batteries and electronics, and new materials.
• The primary tool for identifying these molecules is tandem mass spectrometry coupled with liquid chromatography (LCMS), which generates a spectrum or "molecular fingerprint" for each molecule.
• The challenge lies in interpreting these spectra; over 90% of them are "dark data" that cannot be matched to known molecular structures.

2. Dreams: An AI to Decode Molecular Spectra

• Researchers developed an AI called Dreams, a neural network trained using self-supervised learning, to interpret molecular spectra.
• Dreams was trained on 201 million unlabeled spectra from the GNPS database.
• The AI learns the "grammar" or "language" of how molecules break apart and are mapped onto spectra, enabling it to infer the chemical and structural properties of the molecules behind the spectra.
• This process is analogous to learning a language by reading many books without explicit definitions.

3. The Dreams Atlas: Mapping the Molecular Landscape

• After training, Dreams decodes the 201 million spectra and maps each molecule onto a multi-dimensional space called the Dreams Atlas.
• The position of each molecule in the atlas is based on its similarity to other molecules; molecules with similar properties are located closer together.
• The Dreams Atlas is similar to how words are mapped in large language models, where words with related meanings are clustered together.
• The atlas serves as a framework or "Wikipedia" for molecules, allowing researchers to find connections and similarities between known and unknown molecules.

4. Key Findings and Applications of the Dreams Atlas

• Connectivity: The Dreams Atlas reveals meaningful connections between molecules, even those that were previously unidentified.
• Novelty Detection: The atlas can determine how "new" a molecule is by its distance from known molecules, indicating potential for novel drugs or materials.
• Food Taxonomy: When applied to spectra from various food items, the Dreams Atlas accurately clustered them according to basic food taxonomy (plant-based, animal-based, beverages), demonstrating the AI's ability to classify molecules based on their properties.
• Psoriasis and Fungicides: The atlas revealed a close linkage between psoriasis and the fungicide esoxystrobin, suggesting a potential correlation between exposure to the chemical and the skin condition.
• Plant Metabolites: The atlas identified a specific plant metabolite shared among seemingly unrelated plant species.
• Cancer and Lipids: A family of lipids was found to be closely associated with type 2 diabetes, brain cancer, lung cancer, and renal cancer.

5. Fine-Tuning Dreams for Specific Predictions

• Dreams can be fine-tuned to predict specific properties of molecules based on their spectra.
• Lipinski's Rule of Five: Dreams was fine-tuned to predict the relevance of a molecule to Lipinski's Rule of Five, identifying potential drug candidates.
• Fluorine Detection: Dreams was fine-tuned to predict the presence of fluorine in a molecule, achieving a 91% precision rate, significantly higher than previous methods (e.g., Sirius at 51%). This is important because fluorine-containing molecules are often very stable and used in pharmaceuticals, non-stick coatings, and electronics.

6. Future Potential and Open Source Availability

• The Dreams Atlas and AI model can be used to discover new drug candidates, anti-aging compounds, molecules for fighting pollution, and molecules for digesting plastic.
• The ultimate goal is to train an AI that can predict the full structure of a molecule from its spectrum.
• The code for Dreams is available on GitHub and Hugging Face under the MIT license, allowing researchers to use and fine-tune the model.

7. Monica AI Assistant

• Monica is an AI assistant that provides access to various AI tools, including GPT, Deepseek, Gemini, Flux, Stable Diffusion, Cling, and High Law.
• It can be used as a browser extension on desktop or mobile devices.
• Monica can summarize articles, generate mind maps, summarize YouTube videos, and generate podcasts.

8. Conclusion

The Dreams AI and Dreams Atlas represent a significant step forward in our ability to understand and utilize the vast, unexplored world of natural molecules. By leveraging self-supervised learning and mass spectrometry data, this technology can uncover hidden connections, predict molecular properties, and accelerate discoveries in various fields, including medicine, materials science, and environmental science. The open-source availability of Dreams further empowers researchers to explore and expand upon this groundbreaking work.