AI for scientific discovery

See:

AI/Deep learning/Neural ODEs

AI/Knowledge representation and reasoning#Applications to medicine and healthcare

Resources

Applications in medicine

AI for Medicine | The Batch | AI News & Insights
MedLLMsPracticalGuide - A curated list of practical guide resources of Medical LLMs (Medical LLMs Tree, Tables, and Papers)
Inteligencia Artificial en la Medicina | IBM
Elevating Healthcare Documentation With Generative AI (opendatascience.com)
Datasets:
- abachaa/Existing-Medical-QA-Datasets: Multimodal Question Answering in the Medical Domain: A summary of Existing Datasets and Systems (github.com)
- https://github.com/Andy-jqa/biomedical-qa-datasets
  - Español: HEAD-QA | A Healthcare Dataset for Complex Reasoning (aghie.github.io)

Talks

#TALK Using AI to Accelerate Scientific Discovery (Demis Hassabis, Deepmind)

Code

#CODE Gt4sd-core (IBM) - GT4SD, an open-source library to accelerate hypothesis generation in the scientific discovery process
#CODE Deep Search
- https://ds4sd.github.io/
- https://research.ibm.com/interactive/deep-search/
- Deep Search extracts and structures data from documents in four steps: Parse, Interpret, Index, and Integrate
- Handling Scientific Articles with Deep Search

Events

References

AI for physical sciences

AI for Earth sciences

#PAPER Deep learning and process understanding for data-driven Earth system science (Reichstein, 2019)
- http://www.ccpo.odu.edu/~klinck/Reprints/PDF/reichsteinNature2019.pdf
- ML approaches are increasingly used to extract patterns and insights from the ever-increasing stream of geospatial data, but current approaches may not be optimal when system behaviour is dominated by spatial or temporal context
- Earth system data are exemplary of all four of the ‘four Vs’ of ‘big data’: volume, velocity, variety and veracity
- One key challenge is to extract interpretable information and knowledge from this big data, possibly almost in real time and integrating between disciplines
- Two major tasks in the coming years: (1) extracting knowledge from the data deluge, and (2) deriving models that learn much more from data than traditional data assimilation approaches can, while still respecting our evolving understanding of nature’s laws
- #TALK Machine-learning-model-data-integration for a better understanding of the Earth System (Markus Reichstein)
#PAPER Machine learning for weather and climate are worlds apart (Watson-Parris 2021)
#PAPER Artificial Intelligence Revolutionises Weather Forecast, Climate Monitoring and Decadal Prediction (Dewitte 2021)
#PAPER Towards neural Earth system modelling by integrating artificial intelligence in Earth system science (Irrgang 2021)
#PAPER Evolution of machine learning in environmental science—A perspective (Hsieh 2022)
#PAPER Machine Learning in Weather Prediction and Climate Analyses—Applications and Perspectives (Bochenek 2022)

AI for medicine

#PAPER OpenMedIA: Open-Source Medical Image Analysis Toolbox and Benchmark under Heterogeneous AI Computing Platforms (Zhuang 2022)
- #CODE https://git.openi.org.cn/OpenMedIA
- OpenMedIA is an open-source toolbox library containing a rich set of deep learning methods for medical image analysis under heterogeneous AI computing platforms
#PAPER Discovery of a structural class of antibiotics with explainable deep learning (2023)
#PAPER Large language models in medicine: the potentials and pitfalls (2023)
#PAPER Paper page - ChatDoctor: A Medical Chat Model Fine-tuned on LLaMA Model using Medical Domain Knowledge (huggingface.co)
#PAPER MEDITRON-70B: Scaling Medical Pretraining for Large Language Models (2023)
- https://github.com/epfLLM/meditron
#PAPER Ascle: A Python Natural Language Processing Toolkit for Medical Text Generation (2023)
#PAPER Towards Adapting Open-Source Large Language Models for Expert-Level Clinical Note Generation (2024)
#PAPER Me LLaMA: Foundation Large Language Models for Medical Applications (2024)
#PAPER Apollo: An Lightweight Multilingual Medical LLM towards Democratizing Medical AI to 6B People (2024)
- https://github.com/FreedomIntelligence/Apollo
#PAPER PMC-LLaMA: toward building open-source language models for medicine | Journal of the American Medical Informatics Association | Oxford Academic (2024)
#PAPER Capabilities of Gemini Models in Medicine (2024)
- Med-Gemini and Meditron: Google and Meta present new LLMs for medicine (the-decoder.com)
#PAPER Accurate structure prediction of biomolecular interactions with AlphaFold 3 (2024)
- Google DeepMind and Isomorphic Labs introduce AlphaFold 3 AI model
- DeepMind's AlphaFold 3 Enhances 3D Biomolecular Modeling
- AlphaFold 3 is a triumph of machine learning. It extends the utility of the previous version beyond proteins, and it computes with unprecedented accuracy how biological molecules will combine, allowing for a more comprehensive understanding of how drugs interact with the body
#PAPER Towards Conversational Diagnostic AI (2024)
- AMIE, a Chatbot That Outperforms Doctors in Diagnostic Conversations
#PAPER #REVIEW A Survey of Large Language Models in Medicine: Progress, Application, and Challenge (2024)
#PAPER #REVIEW The advancement of artificial intelligence in biomedical research and health innovation: challenges and opportunities in emerging economies (2024)
#PAPER #REVIEW Unveiling Recent Trends in Biomedical Artificial Intelligence Research: Analysis of Top-Cited Papers (2024)
#PAPER A generalist vision–language foundation model for diverse biomedical tasks | Nature Medicine (2024)
#PAPER Med42-v2: A Suite of Clinical LLMs (2024)
#PAPER #REVIEW The application of large language models in medicine: A scoping review (2024)
#PAPER #REVIEW Generative artificial intelligence in healthcare from the perspective of digital media: Applications, opportunities and challenges (2024)
- Applications of Generative AI
  - Medical Advice and Diagnosis: AI provides initial medical recommendations and triage guidance.
  - Mental Health Support: Chatbots assist with therapeutic conversations, available 24/7.
  - Drug Discovery: AI accelerates drug development and enables personalized medicine.
  - Training and Education: Virtual simulations enhance medical training.
  - Data Synthesis: AI generates synthetic medical data for model training.
- Opportunities
  - Improved Accessibility: Expands healthcare access in underserved regions.
  - Cost Efficiency: Reduces operational expenses through task automation.
  - Enhanced Diagnostics: Detects patterns in medical data for earlier diagnosis.
  - Operational Efficiency: Streamlines workflows, allowing professionals to focus on care.
  - Scalability: Supports healthcare delivery to larger populations effectively.
- Challenges
  - Accuracy and Reliability: Risks of incorrect or misleading AI-generated advice.
  - Ethical Concerns: Issues with consent and reliance on machine decisions.
  - Privacy Risks: Sensitive data is vulnerable to breaches or misuse.
  - Bias: Risk of perpetuating inequities from biased training datasets.
  - Regulation: Navigating complex legal landscapes remains a barrier.
- Interconnections
  - Applications and opportunities are deeply linked to challenges, e.g., enhanced accessibility via AI also raises questions about reliability and quality. Careful balancing is needed to leverage benefits while mitigating risks.
- Recommendations
  - Ethical Frameworks: Develop guidelines for responsible AI use.
  - Transparency: Focus on interpretable AI models for trustworthiness.
  - Collaborations: Foster partnerships between AI experts, healthcare providers, and policymakers.
  - Public Awareness: Educate stakeholders about AI's capabilities and limitations.
  - Bias Mitigation: Invest in strategies to promote fairness and equity in AI applications.
#PAPER Empowering biomedical discovery with AI agents (2024)
- The study envisions AI agents are designed to collaborate with human researchers, combining human creativity and expertise with AI’s capacity to analyze extensive datasets, explore hypothesis spaces, and perform repetitive tasks.
- Key features of these AI agents include:
  - Skeptical Learning and Reasoning: The ability to critically assess and refine hypotheses and experimental designs.
  - Structured Memory for Continual Learning: Utilizing large language models and generative models to maintain and update knowledge bases, facilitating ongoing learning.
  - Integration of Scientific Knowledge: Incorporating existing biological principles and theories to inform decision-making processes.
- Areas with potential applications of AI agents: hybrid cell simulation, programmable control of phenotypes, design of cellular circuits, and development of new therapies.
- By fostering collaboration between AI systems and human researchers, the authors aim to accelerate biomedical discoveries and address complex biological challenges.
#PAPER A foundation model for clinician-centered drug repurposing | Nature Medicine (2024)
- Marinka Zitnik - Graph Neural Networks for Biomedical Data - YouTube
#PAPER #REVIEW A Comprehensive Review of Generative AI in Healthcare (2024)