AI for Science Get a sneak peek at new achievements, new developments, and new perspectives——

* DeepMind's latest research FunSearch published in Nature

* Google launches MedLM, a healthcare industry model

* Jingtai Technology sprints to the Hong Kong Stock Exchange, AI + robots empower AI for Science

* GHDDI reaches cooperation with Microsoft Research Center for Scientific Intelligence

* AI tools for seismological processing and analysis open source

* my country's first independently developed Earth system model announced to be open source

* Baidu PaddlePaddle team builds HelixDock, a protein-small molecule docking conformation prediction model

* Domestic research team publishes carbon emission prediction method and system based on hybrid machine learning

* Apple chip "exclusive customized version" machine learning framework open source

More details below~

Company News

DeepMind  Latest research FunSearch published in Nature

Google DeepMind's latest research FunSearch is a method for searching for new solutions in mathematics and computer science. FunSearch works by pairing a pre-trained large model (LLM) with an automatic "evaluator". The former aims to provide creative solutions in the form of computer code, while the latter is responsible for preventing hallucinations and incorrect ideas. Through iterations back and forth between these two components, the initial solution "evolves" into new knowledge. FunSearch discovered new solutions to the upper bound set problem, a long-standing unsolved problem in mathematics, representing the first time that a challenging open problem in science or mathematics has been discovered using a large model.Paper address:nature.com/articles/s41586-023-06924-6

Google launches MedLM, a healthcare industry model

Recently, Google announced the launch of a new set of AI models for healthcare, MedLM, which is designed to help clinicians and researchers conduct complex research, summarize doctor-patient interactions, etc. This move marks Google's latest attempt to monetize AI tools in the healthcare industry and is also an important milestone in the digital transformation of the healthcare industry. First, MedLM can help clinicians and researchers conduct complex research and data analysis, and improve the accuracy and efficiency of medical diagnosis. Secondly, MedLM can summarize doctor-patient interactions and provide doctors with better patient management and service experience. In addition, MedLM can also provide healthcare institutions with better data management and analysis tools to improve the efficiency of medical resource utilization.

Jingtai Technology is sprinting towards the Hong Kong Stock Exchange, AI+Robots empower AI for Science

QuantumPharm Inc. (QuantumPharm) officially submitted its prospectus to the Hong Kong Stock Exchange last month, intending to list on the main board under Rule 18C. Rule 18C is mainly aimed at special technology companies, and has high requirements for the technological attributes of the industry, involving new generation information technology, advanced hardware and software, advanced materials, new energy and energy conservation and environmental protection, new food and agricultural technology and other industry fields. QuantumPharm is one of the few pharmaceutical and material science R&D companies in the world that has first-principles computing based on quantum physics, advanced artificial intelligence technology and automated wet laboratory capabilities. It is also one of the few pharmaceutical and material science R&D platforms driven by quantum physics + AI + automation in the world.

GHDDI and Microsoft Research Center for Scientific Intelligence Collaborate

Recently, the Global Health Drug Discovery Institute (GHDDI) and Microsoft Research AI4Science announced a collaboration to jointly develop generative artificial intelligence and basic large-scale model technologies in the field of global health infectious diseases, focusing on on-site transformation and accelerating the development of innovative drugs. Previously, the two parties have successfully designed a variety of small molecule inhibitors with new structures in the study of Mycobacterium tuberculosis and key target proteins of coronaviruses.

BioGeometry and Zhipu AI jointly build natural language-life languageMultimodal large model

Beijing Biogeometry Biotechnology Co., Ltd. and Beijing Zhipu Huazhang Technology Co., Ltd. recently announced a strategic partnership to jointly build a natural language-life language multimodal large model. The model is expected to enhance the practicality of generative artificial intelligence platforms in the fields of life science and medical research.

Tools and Resources

AI tool for seismological processing and analysis is open source

Open source tools for seismological processing and analysis, currently including: phase picking, polarization, and dispersion extraction. The tool has open sourced the 100Hz model for China, some of which are trained based on the CSNCD dataset, and the PgSgPnSn four phase picking models have the highest accuracy.accessaddress:https://gitee.com/cangyeone/seismological-ai-tools

my country's first independently developed Earth system model announced to be open source

Recently, the Institute of Atmospheric Physics of the Chinese Academy of Sciences released my country's first "complete" Earth system numerical model with independent intellectual property rights and announced the release of its source code. This model includes a complete climate system and ecological environment system, integrating eight subsystem models such as atmospheric circulation and ocean circulation. It is also a major national scientific and technological infrastructure.Earth System Numerical Simulation Facility"The core software of the program, totaling about 2.7 million lines of program code, is called "Earth Laboratory".

Baidu Propeller Team builds HelixDock, a protein-small molecule docking conformation prediction model  

The Baidu PaddlePaddle team built the protein-small molecule docking conformation prediction model HelixDock by constructing a large-scale simulation data set and upgrading geometry-based neural networks, which greatly improved the accuracy of conformation prediction.For more results, see the HelixDock article:https://arxiv.org/abs/2310.13913
Propeller access address:https://paddlehelix.baidu.com/

Domestic research team publishes carbon emission prediction method and system based on hybrid machine learning

A domestic research team has disclosed a carbon emission prediction method and system based on hybrid machine learning. The method processes the data set through a target combination model to obtain the carbon emission prediction results. Among them, the target combination model realizes the optimal weighted combination of the univariate time series prediction and the multivariate driving factor model through target calculation weights, taking into account the advantages of each model and improving the accuracy of carbon emission prediction.Visit URL:https://cprs.patentstar.com.cn/Search/Detail?ANE=9HFF9IBA9GDC5BCA8GBA9FHE9AHA8BCA9DFB9CFF9GFF7BDA

Apple's custom-made machine learning framework is now open source

MLX is a chip designed specifically for AppleMachine Learning Framework (Click here for detailed explanation), aims to support efficient training and deployment of models on Apple chips while ensuring user-friendliness. Its design concept is simple, referring to frameworks such as NumPy, PyTorch, Jax, and ArrayFire, including key functions such as lazy computation and dynamic graph construction.Visit URL:https://github.com/ml-explore/mlx/tree/main/examples

Research resultsDANTE: Towards large-scale optoelectronic intelligent computing 

Training large-scale optoelectronic neural networks with dual-neuron optical-artificial learning

*  source：Nature Communications

*  field：Neural network, optoelectronic intelligence

*  author：Fang Lu's research group, Department of Electronic Engineering, Tsinghua University

The research team proposed an optical-artificial dual neuron learning architecture (DANTE) for large-scale optoelectronic intelligent computing. The optical neurons accurately model the light field calculation process, the artificial neurons use lightweight mapping functions to establish jump connections to assist gradient propagation, and the global artificial neurons and local optical neurons are iteratively optimized with an alternating learning mechanism. While ensuring the effectiveness of learning, it greatly reduces the spatiotemporal complexity of training, making it possible to train larger and deeper optoelectronic neural networks.

Read the original article:https://www.nature.com/articles/s41467-023-42984-y

Convolutional Neural Network Framework PtyNet: Synchrotron Radiation Massive Data Processing

An efficient ptychography reconstruction strategy through fine-tuning of large pre-trained deep learning model

*  source:iScience

*  field: Data mining, convolutional neural networks

*  author：Chinese Academy of Sciences Team

The research team developed a convolutional neural network framework called PtyNet to recover the accurate projection of objects from X-ray Ptychography experimental data. With the support of a powerful computing cluster, PtyNet can quickly obtain data from synchrotron radiation sources for training and quickly reconstruct images of users' experimental data.

Read the original article:https://doi.org/10.1016/j.isci.2023.108420

Predicting multiple conformations through sequence clustering and AlphaFold2

Predicting multiple conformations via sequence clustering and AlphaFold2

*  source：Nature

*  field：Bioinformatics

*  author: Research team from Brandeis University and Howard Hughes Medical Institute, Harvard University and Cambridge University

The research teamSequence similarityClustering of multiple sequence alignments (MSAs) enables AF2 to sample alternate states of known metamorphic proteins with high confidence. At the same time, the researchers used the AF-Cluster method to study the evolutionary distribution of the predicted structure of the metamorphic protein KaiB5 and found that the predictions for both conformations were distributed in clusters of the KaiB family.

Read the original article:

https://www.nature.com/articles/s41586-023-06832-9

ProRefiner: a model for inverse protein folding design

ProRefiner: an entropy-based refining strategy for inverse protein folding with global graph attention

*  source：Nature Communications

*  field：Biological genes, deep learning

*  author：Research teams from the Chinese University of Hong Kong, Zhijiang Laboratory, Huawei Noah's Ark Laboratory and Nanjing Medical University

The research team introduced ProRefiner, a memory-efficient global graph attention model that can fully exploit the denoising context, and demonstrated the applicability of ProRefiner in redesigning transposon-associated transposase B (TnpB), with 6 out of 20 proposed variants showing improved gene editing activity.

Read the original article:https://www.nature.com/articles/s41467-023-43166-6

KPGT: A self-supervised learning framework

A knowledge-guided pre-training framework for improving molecular representation learning

*  source：Nature Communications

*  field: Biomolecules, Drug Discovery

*  author：Research team from Tsinghua University, Westlake University and Zhijiang Laboratory

The research team proposed Knowledge-guided Pre-training of Graph Transformer (KPGT), a self-supervised learning framework that provides improved, generalizable, and robust molecular property predictions through significantly enhanced molecular representation learning. The KPGT framework integrates a graph transformer designed specifically for molecular graphs and a knowledge-guided pre-training strategy to fully capture the structural and semantic knowledge of molecules.Read the original article:https://www.nature.com/articles/s41467-023-43214-1

Event Review

CoRL Conference Concludes, Best Paper and Best System Paper Announced

The 2023 Conference on Robot Learning (CoRL) was held in Atlanta, USA last month. According to official data, 199 papers from 25 countries were selected for CoRL this year, with popular topics including manipulation and reinforcement learning.

The best paper award went to "Distilled Feature Fields Enable Few-Shot Language-Guided Manipulation"

*  author：William Shen, Ge Yang, Alan Yu, Jensen Wong, Leslie Pack Kaelbling, Phillip Isola

* mechanism：MIT CSAIL、IAIFI

Read the original article:https://openreview.net/forum?id=Rb0nGIt_kh5
For other awards, please visit the official website:https://www.corl2023.org/awards

NASSMA 2022 AI4Science Workshop Topics

The seminar was jointly organized by NASSMA, Mohammed VI Polytechnic University, Google Deepmind and other institutions. Currently, the seminar presentation PPT and live broadcast replay are online.

The above is all the content to be shared in this issue of "Science AI Weekly"~

If you have the latest research results, first-hand information about companies, etc. about AI for Science, please leave a message "Revelation".