Agent & Prediction

Intelligent agents for chemical and biological research

Experiment verification agent

• 7x24 hour compliance gatekeeper: using ELN's audit trail and timestamp data, the agent promptly checks the completeness of experimental records, automatically identifies potential compliance risks, and ensures compliance with FDA/NMPA requirements.

Bayesian optimization agent

• Even a novice can design expert-level formulations: In the InTable module of ELN, the agent guides you through a conversation to input experimental goals (e.g., maximum yield) and constraints, automatically running algorithms to recommend optimal experimental designs. Unlike traditional DOE, which requires pre-planning and running a large matrix of experiments before analysis, this dynamic iterative approach adjusts in real time based on new results, typically saving 30%-50% of experimental effort and converging quickly to the best outcome with minimal cost.

Structure-activity relationship extraction agent

• Extracts wisdom from papers and patents: Inpaper agent widely supports mainstream file formats including pdf, word and excel. It deeply parses external literature and internal reports, penetrates complex document layouts, and automatically extracts key compound names, molecular structure descriptions, and activity data (e.g., IC50, EC50), and assembles structure-activity relationships (SAR)—turning "dead data" trapped in various document formats into computable, searchable "living knowledge".

Project monitoring agent

• All-weather project risk controller: In AtomDeep Project, the agent analyzes project tasks and related experiment progress in real time. You can ask at any time: "Is the project delayed?" "Is the cost over budget?" The AI can comprehensively reason and provide early warnings.

ADMET prediction

• Know success or failure without moving: This function is integrated into the InDraw client and provided by partner Hangzhou Carbon Silicon Intelligence. The agent transforms drug discovery from the traditional "synthesize first, test later" approach into an efficient "predict first, synthesize later" model. Simply select a structure on the canvas, and the agent quickly calculates and evaluates about 90 key properties covering five major categories: absorption, distribution, metabolism, excretion, and toxicity (ADMET). Acting like a virtual screener, it helps you accurately eliminate molecules with poor druggability at an early stage, significantly reducing the cost and risk of trial and error in later experiments.

pKa prediction

• Microscopic insight, precise characterization: Integrated into the InDraw client and built on the industry-leading Uni-pKa open-source model, it not only provides macroscopic pKa but also delivers microscopic pKa that reflects the dissociation ability of specific atomic sites, while also covering the pKa of conjugate acids of bases. It offers you more accurate and comprehensive ionization state predictions, helping you precisely control molecular behavior in different pH environments, and can be used to predict drug molecule solubility, membrane permeability, and protein binding capacity.

Chemical structure image recognition

• Self-developed OCSR engine, accurately restoring the chemical world: Deeply integrated into InDraw and InPaper, it is powered by Ingle's self-developed OCSR model. It can accurately identify structural formulas and reaction formulas in images, instantly converting pixels into editable structures, while also supporting efficient batch recognition services. The accuracy of chemical structure recognition has exceeded 95%, significantly outperforming mainstream open-source products such as MolNextR and MolScribe, which achieve less than 90%, ensuring the accuracy and reliability of data at the source.

3D simulation and molecular docking

• From plane to 3D, visualize molecular docking: Integrated into InDraw, it automatically calculates the lowest energy conformation of a molecule, instantly converting a flat chemical structure into a vivid 3D model. Going further, it supports importing complex crystal complexes, allowing you to intuitively view the docking status of ligands and proteins, and examine key interactions such as hydrogen bonds and van der Waals forces from multiple angles, helping you validate the rationality of drug designs.