TL;DR
We discuss recent reports that GPT can play chess, followed by a brief overview of a recent article that introduces a drug off-target interaction discovery methodology.
GPT and Chess
A recent tweet made the observation that GPT3.5 is capable of playing chess with 1800 ELO. This corresponds to the strength of a strong club player (source). A number of online commentators took the tweet to mean that GPT is capable of reasoning. I don’t personally believe this to be the case. 1800 ELO is fairly weak as far as chess agents go (Stockfish has an ELO of about 3550). It is unlikely that GPT is achieving zero-shot performance on chess; while the training data is secret, it would be very surprising to me if there were not a sizable amount of chess related training data. GPT is not specifically trained to play chess, so it does a decent but not great job at chess when compared with more specialized training agents. None of this is surprising behavior.
GPT’s strength is that it does reasonably well at a broad range of tasks. This is indeed a powerful and unexpected capability that arguably no past model has achieved. GPT appears to generate generalizable language embeddings and has learned some basic ability to transform on those embeddings. But at least for now, GPT persistently underperforms more specialized models and algorithms for most tasks. For example, GPT still cannot add or multiply numbers correctly without access to an external mathematical engine. It is also worth noting that Google search has an even broader range of “knowledge” it can access. GPT takes one (big) step further to make a cogent language-based interface, but fundamentally I don’t believe it has yet shown the ability to reason. This may well change in the next few years, but it’s important for us to be as precise as we can about model capabilities. Hype and AI doom worries have the potential to choke off societally useful advances.
Identifying Intracellular Drug Targets Using a Metabolomics-guided Framework
Drug development efforts design compounds to bind to specific targets associated with a disease. However, drugs may also interact with unintended targets (off-target effects), leading to undesired effects. The goal of drug off-target discovery is to identify these unintended interactions and understand how they may affect the overall efficacy and safety of the drug.
This recent article focuses on understanding intracellular drug targets, particularly in the context of identifying off-target interaction of a multi-valent antibiotic compound, CD15-3, which was designed to target dihydrofolate reductase (DHFR). The authors have developed a workflow that combines machine learning, metabolic modeling, and protein structure analysis to identify potential off-targets from metabolomics data as illustrated in the diagram below. Using their method, they identify HPPK as a secondary target of CD15-3.
Systematic methods to discover off-targets could prove very powerful for understanding side-effects and toxicity. The current study only examines one cell type, but it would be useful to extend this method to a bank of cell types to understand the effect of a proposed drug across the body. In the future, such analyses could possibly become standard requirements for drug approvals.’
Interesting Links from Around the Web
https://www.anandtech.com/show/20066/intel-highna-lithography-update-dev-work-on-intel-18a-production-in-future-node: Intel 18a is moving faster than expected, so high-NA EUV may be used for the subsequent node (currently code named Intel Next).
https://www.nature.com/articles/d41586-023-02984-w: Questions remain about AlphaFold’s effectiveness for drug discovery.
https://www.nextplatform.com/2023/09/22/intel-gets-its-chiplets-in-order-with-5th-gen-xeon-sps/, https://www.nextplatform.com/2023/09/19/intel-xeon-roadmap-on-track-288-core-sierra-forrest-coming-soon/ : Intel is working to improve its server processor offerings. Notably, Intel will ship a 288 core server.
https://www.science.org/doi/10.1126/scirobotics.adg6042: A two-armed endoscopic robot for neurosurgeons.
Feedback and Comments
Please feel free to email me directly (bharath@deepforestsci.com) with your feedback and comments!
About
Deep Into the Forest is a newsletter by Deep Forest Sciences, Inc. We’re a deep tech R&D company building Chiron, an AI-powered scientific discovery engine for the biotech/pharma industries. Deep Forest Sciences leads the development of the open source DeepChem ecosystem. Partner with us to apply our foundational AI technologies to hard real-world problems in drug discovery. Get in touch with us at partnerships@deepforestsci.com!
Credits
Lead Author: Bharath Ramsundar, Ph.D.
Editor: Sandya Subramanian, Ph.D.
Research and Writing: Rida Irfan