Earthquake in Science! Self-Made PhD Publishes Six Consecutive Nature Papers — AI-Powered Protein Design Silences All Doubts

On May 8, 2024, Google DeepMind and Isomorphic Labs jointly released the latest artificial intelligence model AlphaFold 3 in the field of proteins in the journal Nature! This model can accurately predict the structures of biomolecules such as proteins, DNA, RNA, and ligands, as well as their interaction patterns. This is another significant breakthrough following AlphaFold 2.

In predicting drug interactions, AlphaFold 3 has achieved unprecedented accuracy, including the binding of proteins with ligands and antibodies with their target proteins. In the PoseBusters benchmark, AlphaFold 3's accuracy is 50% higher than the best existing traditional methods, and it requires no structural information input, making it the first artificial intelligence system to surpass traditional physics-based prediction tools. This ability to predict antibody-protein binding is crucial for understanding various aspects of human immune responses and designing new antibodies.

Proteins are the building blocks of life. From constructing our body tissues, regulating metabolism, to fighting diseases, proteins are everywhere. However, proteins in nature do not always meet our needs. Traditional protein engineering methods often rely on trial and error and experience, which are time-consuming, labor-intensive, and inefficient. The emergence of AI has opened a new door for us. Through machine learning and deep learning algorithms, AI can quickly analyze massive amounts of protein data, predict protein structures, and even design entirely new proteins from scratch.

As the most anticipated technology of 2024!

There are very few resources and learning pathways for AI protein design, making specialized training urgently needed! Zhengzhou Qingrui Information Technology Co., Ltd., in collaboration with Tsinghua University, Peking University, Westlake University, Zhejiang University, University of Science and Technology of China, Tianjin University, and the Institute of Materia Medica, Chinese Academy of Medical Sciences, has held over forty training sessions with more than 5,000 participants! The training has received extremely high praise from attendees, some of whom have published in top international journals such as Nature, Cell, and Science!

*Operations involving the use of code/computational tools

1. Basic Concepts

1.1. Introduction to Machine Learning: From Handwritten Digit Recognition to Large Language Models

1.2. Review of Protein Structure Prediction and Design

1.3. Introduction to Linux

1.4. Code Environment: VS Code and Jupyter Notebook*

1.5.Introduction to Key Python Concepts*

2. Commonly Used Methods for Analyzing/Visualizing Proteins and Related Molecules

2.1. Common Databases and Homologous Sequence Search and MSA Construction

2.2. Visualizing Protein Structures Using PyMOL and Mol*

2.3. Using Biopython and Biotite to Analyze Biological Sequence and Structure Data*

2.4. Using fpocket and point-site to analyze protein structure pockets*

3. Differences Between Deep Learning Protein Design and Traditional Protein Design

3.1. The Essence of Deep Learning

3.2. Traditional Methods: All-Atom Energy Function Rosetta and Statistical Potentials

3.3. Deep Learning: Geometric Deep Learning

3.4. Complementarity of Deep Learning and Traditional Physical Methods

3.5. Superiority of Deep Learning in Protein Design

4. Protein Language Model

4.1. Language Models: From RNN to Transformers

4.2. Understanding the Protein Language

4.3. Generative Protein Language Models

4.4. Comparative Analysis of Structural Models and Language Models

5. Protein Function and Property Prediction Based on Deep Learning

5.1. Protein Function Classification Prediction*

5.2.Signal Peptide, Transmembrane Region, Subcellular Localization Prediction

5.3. Protein Homologous Structure Search

5.4. Enzyme Active Site Prediction

1. Traditional Protein (Complex) Structure Prediction

1.1. Homology modeling using Modeller and Swiss-Model

1.2. Ab initio modeling based on molecular dynamics

1.3. Molecular docking including non-protein parts, practical operation of AutoDock Vina*

2. Modern Deep Learning for Protein Structure Prediction

2.1 RaptorX: From Computer Vision to Protein Structure

2.2.AlphaFold2

2.3.AlphaFold3: Generative Structure Prediction

2.4.ESMFold: The Integration of Language Models and Structure Prediction

3. AlphaFold2 Principle Review

3.1. From Co-evolution to Structure

3.2. Attention Mechanism

3.3.EvoFormer

3.4.Structural Module

4.Introduction to AlphaFold3

4.1. Diffusion Model

4.2. Training Data

4.3. Achievements and Shortcomings of AlphaFold3

5. Practical Operation and Result Analysis of AlphaFold2/3

5.1. AlphaFold2 Practical Operation*

5.2.AlphaFold2 Analysis*

5.3. Using the AlphaFold Server*

5.4. Local Version of AlphaFold3*

5.5.AlphaFold3 Analysis*

6.ESMFold

6.1. From Language Models to Structural Prediction

6.2. When to Use ESMFold and When to Use AlphaFold

6.3.ESMFold Usage*

1. Traditional Protein Sequence Design

1.1. Based on All-Atom Force Field*

1.2. Based on Statistical Potential

2. Protein Sequence Design with Structural Knowledge-Infused Language Models

2.1. Introduction to the Principle of ESM-IF

2.2. Application of ESM-IF*

3. CNN-Based Sequence Design

3.1. Introduction to CNN Principles

3.2.DenseCPD Design Method

3.3. Design Methods with Side Chain Conformations

4. GNN-Based Sequence Design

4.1. Analysis of the Success Experience of ProteinMPNN

4.2.Wide Applications of ProteinMPNN

4.3.ProteinMPNN Practical Operation*

5. Other Sequence Design Models

5.1. Introduction to ABACUS-R and Practical Operation*

5.2.CarbonDesign: From Structure Prediction to Sequence Design*

5.3.CARBonAra Environment-Aware Sequence Design*

6. Application of Fixed Main Chain Sequence Design in Functional Protein Design

6.1. Optimization of New Scaffold Protein Expression (Reproduction of Science Article)*

6.2. Antibody Affinity Optimization (Science Article Reproduction)*

6.3. Comprehensive Optimization of Enzyme Properties Combined with Evolutionary Information (Reproduction of JACS Article)*

1. Review of Traditional Thinking

1.1. Domain Splicing

1.2.SCUBA: Side-Chain-Free Protein Force Field

2. Binder Design Based on Geometric Deep Learning of Protein Surfaces

2.1. Introduction to MASIF Principle

2.2. masif for Identifying PPI Hotspots on Protein Surfaces

2.3.Masif Design Binder

3. Protein Backbone Design Model Based on Diffusion Model

3.1.FrameDiff: IPA-Based Backbone Generation*

3.2.Chroma: Design of Equivariant Graph Neural Network Architecture

3.3.RFDiffusion: Based on RosettaFold

3.4.RFDiffusion-All-Atom: Based on RosettaFold-All-Atom

4. Sequence-Structure Co-design

4.1.trDesign

4.2.AlphaFold Hallucination

4.3.Rfjoint

4.4.Protein Generator

5. De novo design of binding proteins Nature Communication article process*

5.1. Selection of Functional Epitopes

5.2. Skeleton Generation with Restrictions

5.3. Iterative Optimization

6. De Novo Design of Luciferase Structure

6.1. Theozyme Theory Explanation

6.2. Skeleton Generation Strategy

6.3. Active Site Design and Activity Evolution

1.Deep Learning Modeling Methods for Language

1.1.Transformer

1.2.BERT: Bidirectional Encoder Representations from Transformers

1.3.GPT: Generative Pre-trained Transformers

2. Representative of Protein Language Models: ESM

2.1. Model Framework

2.2.ESM Series Work: ESM-1/2, MSA Transformer, ESM3

2.3.ESM Model Practical Operation*

3. Functional Protein Design Based on Protein Language Models

3.1. The Paradigm of Pre-training + Fine-tuning

3.2.Conditional Generative Models: Progen and ZymCTRL

3.3.Progen Case Analysis

3.4. Fine-tuning ZymCTRL*

4. Non-Autoregressive Sequence Generation Model

4.1.ProteinGAN: Generating Sequences

4.2.DeepEvo: Generating Thermostable Enzymes

4.3.Prot-VAE

4.4.P450Diffusion: Designing Functional P450 Based on Diffusion Model*

5. Evaluation Index After Functional Protein Generation

5.1. Natural Sequence Similarity Assessment*

5.2. Diversity Assessment*

5.3. Structural Rationality Assessment*

1. Enzyme Property Prediction

1.1 Introduction to DLKcat and GotEnzyme Databases

1.2 UniKP: Utilizing Pre-trained Models to Mine and Transform Kcat*

1.3.CLEAN: Contrastive Learning-based EC Number Prediction for Mining Rare Dehalogenases*

2. Protein Thermal Stability Engineering

2.1.Introduction to MutCompute

2.2. Engineering PETase with MutCompute (Nature)*

2.3.Introduction and Usage of ThermoMPNN*

2.4. Introduction and Usage of Pythia*

3. Machine Learning-Assisted Directed Evolution/Protein Engineering

3.1. Zero-shot Mutation Effect Prediction Principle

3.2. Zero-shot Remodeling Gene Editing Enzyme*

3.3. Low-N Strategy for Protein Engineering

3.4. Evo-tuning of Pre-trained Models*

3.5.Introduction to ECNet

3.6. Prediction of Mutation Effects in Protein Interactions

4. Based on your own experimental data, train your own neural network*

4.1. Neural Network Training Framework

4.2. Data Collection and Organization

4.3. Feature Extraction Methods

4.4. Selection of Pre-trained Models

4.5. Model Training and Testing

4.6. Prediction of New Mutations

5. Deep Learning-Assisted New Enzyme Mining*

5.1. Gene Editing Deaminase Mining (Reproduction of Cell Work)

5.2. Mining of Thermostable Plastic Hydrolases (Nature Communications)

5.3.UseFoldSeekPerform structure-based mining

Morning

Background, Theoretical Knowledge, and Tool Preparation

1. Introduction and Use of the PDB Database

1.1 Introduction to the Database

1.2 Query and Selection of Target Protein Structures

1.3 Download of Target Protein Structure Sequence

1.4 Download and Preprocessing of Target Proteins

1.5 Batch Download Protein Crystal Structures

2. Introduction and Usage of Pymol

2.1 Introduction to Basic Software Operations and Fundamental Knowledge

2.2 Protein-Ligand Interaction Diagram

2.3 Protein-Ligand Small Molecule Surface Diagram, Electrostatic Potential Representation

2.4 Protein-Ligand Structure Superposition and Alignment

2.5 Plotting Interaction Forces

3.Introduction and Use of Notepad

3.1 Introduction to Advantages and Main Functions

3.2 Interface and Basic Operations Introduction

3.3 Plugin Installation and Usage

Afternoon

General Protein

-Ligand Molecular Docking Explanation

1. Introduction to Relevant Theories of Docking

1.1 The Concept and Basic Principles of Molecular Docking

1.2 Basic Methods of Molecular Docking

1.3 Commonly Used Software for Molecular Docking

1.4 General Process of Molecular Docking

2. Conventional Protein-Ligand Docking

2.1 Collection of Receptor and Ligand Molecules

2.2 Processing of Complex Pre-conformations

2.3 Preparation of Receptor and Ligand Molecules

2.4 Protein-Ligand Docking

2.5 Analysis of Docking Results

Taking the main protease of the新冠病毒 protein and related inhibitors as an example

Virtual Screening

1. Introduction and Download of Small Molecule Database

2. Introduction to Related Programs

2.1 Introduction and Usage of OpenBabel

2.2 Introduction and Usage of ChemDraw

3. Preprocessing for Virtual Screening

4. The Process and Practical Demonstration of Virtual Screening

Case: Screening for Main Protease Inhibitors of the Coronavirus

5. Result Analysis and Plotting

6. Drug ADME Prediction

6.1 Introduction to ADME Concepts

6.2 Introduction to Relevant Websites and Software for Prediction

6.3 Analysis of Prediction Results

Methods for Expanding Docking Usage

1. Protein-Protein Docking

1.1 Application Scenarios of Protein-Protein Docking

1.2 Introduction to Related Programs

1.3 Collection and Preprocessing of Target Proteins

1.4 Calculation Using Examples

1.5 Preset of Key Residues

1.6 Acquisition of Results and File Types

1.7 Analysis of Results

Currently, the popular target

PD-1/PD-L1, etc.

2. Docking involving metalloenzymes

2.1 Background Introduction of Metalloenzyme Protein-Ligand

2.2 Collection and Preprocessing of Proteins and Ligand Molecules

2.3 Treatment of Metal Ions

2.4 Docking of Metal Cofactor Protein-Ligand

2.5 Result Analysis

Taking human farnesyltransferase and its inhibitors as examples

3. Protein-polysaccharide molecular docking

4.1 Protein-Polysaccharide Interactions

4.2 Key Points of Docking Processing

4.3 The Process of Protein-Polysaccharide Molecular Docking

4.4 Protein-Polysaccharide Molecular Docking

4.5 Analysis of Related Results

Taking α-glucosidase and polysaccharide molecular docking as examples

5. Nucleic Acid-Small Molecule Docking

5.1 Application Status of Nucleic Acid-Small Molecules

5.2 Introduction to Related Procedures

5.3 Types of Nucleic Acid-Small Molecule Binding

5.4 Nucleic Acid-Small Molecule Docking

5.5 Analysis of Related Results

Human Telomere

g - Quadruple chain and ligand molecular docking as an example.

Introduction to Operation Process and Practical Demonstration

Methods for Expanding Docking Usage

1.Flexible Docking

1.1 Introduction to the Use Cases of Flexible Docking

1.2 Advantages of Flexible Docking

1.3 Protein-Ligand Flexible Docking

Focus: Method for setting flexible residues

1.4 Analysis of Related Results

Cyclin-dependent kinase

2 (CDK2) with ligand 1CK as an example

2. Covalent Docking

2.1 Introduction to Two Covalent Docking Methods

2.1.1 Flexible Side Chain Method

2.1.2 Two-Point Attractor Method

2.2 Collection and Preprocessing of Proteins and Ligands

2.3 Covalent Docking of Covalent Drug Molecules with Target Proteins

2.4 Comparison of Results

Taking the currently popular covalent drugs for COVID-19 as an example.

3. Protein-Hydration Docking

3.1 The Significance and Methods of Hydration in Protein-Ligand Interactions

3.2 Collection and Preprocessing of Proteins and Ligands

3.3 Preparation of Relevant Parameters for Docking

Key Point: The Addition and Treatment of Water Molecules

3.4 Protein-Water Molecule-Ligand Docking

3.5 Result Analysis

Acetylcholine-binding protein

(AChBP) with nicotine complex as an example

Molecular Dynamics Simulation (Linux and GROMACS Installation and Usage)

1. Introduction and Simple Use of Linux System

1.1 Common Linux Command Lines

1.2 Common Program Installation on Linux

1.3 Experience: How to Perform Virtual Screening on Linux

2. Introduction to Molecular Dynamics Theory

2.1 Principles of Molecular Dynamics Simulation

2.2 Methods and Related Programs of Molecular Dynamics Simulation

2.3 Introduction to Relevant Force Fields

3. Introduction and Usage of Gromacs

Focus: Introduction to Main Commands and Parameters

4. Introduction and Use of Origin

Execution of Solvated Molecular Dynamics Simulations

1. General Process for Handling Solvated Proteins

2. Preparation of Protein Crystals

3. Energy Minimization of Structures

4. Pre-equilibration of the system

5. Unrestricted Molecular Dynamics Simulation

6. Presentation and Interpretation of Molecular Dynamics Results

Taking lysozyme in water as an example

1.Overview of AIDD and Introduction to Comprehensive Drug Database

2. Overview of Artificial Intelligence-Aided Drug Design (AIDD)

3. Installation Environment

(1)anaconda

(2)vscode

(3)pycharm

(4) Virtual Environment

4. Basic Usage of Third-Party Libraries

(1)numpy

(2)pandas

(3)matplotlib

(4)requests

5. Acquisition Methods of Comprehensive Databases for Multiple Drugs

(1) KEGG (requests crawler)

(2)Chebi（libChEBIpy）

(3)PubChem（pubchempy / requests）

(4)ChEMBL（chembl_webresource_client）

(5)BiGG（curl）

(6)PDB（pypdb）

1. Machine Learning

(1) Types of Machine Learning:

① Supervised Learning

②Unsupervised Learning

③ Reinforcement Learning

(2) Typical Machine Learning Methods

①Decision Tree

② Support Vector Machine

③Naive Bayes

④Neural Network

⑤Convolutional Neural Network

(3) Evaluation and Validation of the Model

(4) Classification Evaluation: Accuracy, Precision, Recall, F1 Score, ROC Curve, AUC Calculation

(5) Regression Evaluation: Mean Absolute Error, Mean Squared Error, R2 Score, Explained Variance Score

(6) Cross-validation

2. Basic Usage of sklearn Toolkit

3.Basic Usage of RDKit Toolkit

4. Compound Encoding Methods and Compound Similarity Theoretical Knowledge

5. Project Practice 1: Molecular Screening Based on ADME and Ro5

6. Project 2: Ligand Screening Based on Compound Similarity

7. Project Practice 3: Molecular Clustering Based on Compound Similarity

8. Project Practice 4: Biological Activity Prediction Based on Machine Learning

9. Project Practice 5: Machine Learning-Based Molecular Toxicity Prediction

1. Graph Neural Network

(1) Framework Introduction: PyG, DGL, TorchDrug

(2) Graph Neural Network Message Passing Mechanism

(3) Design of Graph Neural Network Datasets

(4) Practical Tasks for Graph Neural Network Node Prediction, Graph Prediction, and Edge Prediction

2. In-depth Analysis of the Paper: DeepTox: Toxicity Prediction Using Deep Learning

3.Project Practice 1: Molecular Toxicity Prediction Based on Graph Neural Networks

(1) Construction of PyG Graph Dataset from SMILES Molecular Data

(2) Molecular Toxicity Prediction Based on GNN

4. Project Practice 2: Prediction of Protein-Ligand Interactions Based on Graph Neural Networks

(1) Protein molecule visualization, constructing PyG graph dataset

(2) Network construction and interaction prediction based on GIN

1.Natural Language Processing

(1)Encoder-Decoder Model

(2) Recurrent Neural Network RNN

(3)Seq2seq

(4)Attention

(5)Transformer

2.Project Practice 1: Molecular Toxicity Prediction Based on Natural Language

(1) SMILES Molecular Dataset Word Vector Representation Method

(2) Molecular Toxicity Prediction Based on NLP Model

3.Project Practice 2: Prediction of Organic Chemical Reaction Yields Based on Transformer (Prediction of chemical reaction yields using deep learning)

4. In-depth Paper Reading and Code Explanation: "Mapping the space of chemical reactions using attention-based neural networks"

Computer-Aided Drug Design Video Replay

A total of 7 days of video lessons, providing all recorded broadcasts + code materials + PPT software + group Q&A

AIDD Artificial Intelligence Drug Discovery and DesignVideo

A total of 5 days of video lessons, providing all recorded broadcasts + code materials + PPT software + group Q&A.