Orchestrate Agent
Plans, delegates, tracks state, and controls feedback loops.
Full-process agentic drug discovery
From a scientific questionto a synthesis-aware candidate
EasternDawn builds traceable discovery systems that connect scientific agents, generative models, live evidence, and quantitative evaluators.
Manuscript claim
To our knowledge, the first multi-agent drug design framework that integrates and automates end-to-end discovery workflows for both small-molecule and peptide therapeutics.
FROGENTGlobal Context
OOrchestrate
RRetrieve
FForge
GGauge
Connected capabilities
4Collaborating agents
11Discovery capabilities
8Benchmark families
3End-to-end case studies
01 / Flagship platform
FROGENT coordinates the full discovery loop
A high-level goal becomes a task graph. Specialized agents gather evidence, generate candidates, evaluate them, and return diagnostics into a shared context for the next decision.
Retrieve Agent
Gathers evidence from literature, databases, and live sources.
Forge Agent
Generates and optimizes candidates, then plans retrosynthesis.
Gauge Agent
Scores candidates through in silico evaluation and diagnostics.
Shared system memory
Global Context
- Target brief and priors
- Objectives and constraints
- Candidate registry
- Scores and failure diagnostics
- Artifact and report index
A scientific goal with explicit constraints
Target or modality, design objectives, thresholds, prior evidence, and human decisions.
One inspectable record of the campaign
Task graph, evidence, candidates, scores, failure diagnostics, artifacts, and agent handoffs.
A decision-ready scientific package
Ranked candidates, evaluation records, synthesis plans, provenance, and a consolidated report.
01
Define the goal
02
Retrieve evidence
03
Design candidates
04
Evaluate and refine
05
Plan synthesis
06
Deliver the report
02 / Scientific portfolio
One company, a composable discovery system
The portfolio is organized by scientific role. FROGENT orchestrates; design engines generate modality-specific candidates; scientific agents run focused reasoning workflows; Model Context Protocol (MCP) connects each capability.
04FROGENTFull-process orchestration
03Scientific AgentsTask-specific reasoning and review
02Scientific Design EnginesModality-specific generation and optimization
01MCP Capability LayerModels, databases, tools, and evaluators
01Scientific Design Engine
TrioMol2
All-Target Molecular Generation
Pocket-conditioned molecular generation at human-proteome scale.
Small molecule
02Scientific Design Engine
TrioPROTAC
Structure-aware degrader design
Structure-aware reasoning across fixed anchors and a generative middle.
Targeted degradation
03Scientific Design Engine
TrioPep
Structure-feedback peptide design
Peptide generation with structure feedback and docking review.
Peptide
04Scientific Design Engine
TrioIRES
Structure-aware ribonucleic acid (RNA) design
RNA candidate prioritization using activity, topology, and structure evidence.
RNA
05Scientific Design Engine
TrioDNA
Cell-type-aware regulatory deoxyribonucleic acid (DNA) design
Regulatory sequence design with cell-type-aware computational evidence.
DNA
06Scientific Agent
Pio
Evidence-aware peptide optimization
Liability scanning, evidence retrieval, design review, and next-experiment planning.
Peptide optimization03 / Capability surface
Eleven connected capabilities across one scientific workflow
Each capability can operate within FROGENT and can evolve into a dedicated MCP-connected engine or agent as the portfolio expands.
- Shared context
- Persistent project state across agents and stages
- MCP interface
- A common action surface for models, data, and tools
- Human checkpoints
- Oversight at consequential scientific decisions
- 01Information retrieval
- 02File acquisition
- 03Target identification
- 04Molecule generation
- 05Small-molecule docking
- 06Peptide docking
- 07Interaction analysis
- 08Lead optimization
- 09Property prediction
- 10Synthesis planning
- 11Report writing
04 / Research evidence
A platform claim backed by task-level evaluation
The supplied FROGENT manuscript reports comparisons with six progressively capable baselines across eight benchmark families, plus three end-to-end design case studies.
Four agents operate through shared state and one tool interface
The source figure details planning, retrieval, generation, evaluation, feedback, and the MCP-connected model and tool surface.
Open full figure ↗
Eight task families test the connected workflow end to end
FROGENT is shown in green against six progressively capable baselines. Definitions and scoring scales remain task specific.
Open full figure ↗Benchmark atlas8 families
- 01Biomedical knowledge
- 02Known-drug retrieval
- 03Disease-target retrieval
- 04Absorption, distribution, metabolism, excretion, and toxicity (ADMET) prediction
- 05Virtual screening
- 06Binding interaction profiling
- 07De novo molecular design
- 08Retrosynthesis planning
Cardiomegaly and congestive heart failure
Target retrieval, structure acquisition, generation, screening, lead optimization, and retrosynthesis.
Glucagon-like peptide-1 (GLP-1) receptor peptide optimization
A design-evaluate-refine loop for peptide candidates under a consistent in silico protocol.
Carbonic anhydrase inhibitor optimization
Structure analysis, analogue generation, evaluation, and synthesis-path repair using retrieved evidence.
A complete computational trail from disease question to synthesis planning
The published workflow connects target retrieval, structure acquisition, candidate generation, ADMET and docking filters, lead optimization, and retrosynthetic planning.
Open full figure ↗Figure provenance
Figures are reproduced from the user-supplied FROGENT manuscript package. They document computational architecture, benchmarks, and case studies; they do not represent wet-lab validation.
Evidence boundary
Benchmark and case-study statements summarize the supplied manuscript. Computational ranking and docking do not establish experimental efficacy, safety, affinity, or clinical performance.
05 / How we work
Platforms, discovery programs, and scientific partnerships
EasternDawn can deliver an integrated workflow, a focused scientific capability, or a research collaboration built around a defined discovery question.
Agentic discovery programs
Traceable workflows from target context through candidate design, evaluation, synthesis planning, and reporting.
MCP capability deployment
Specialized models, databases, and evaluators packaged as composable tools for scientific agents.
Custom scientific agents
Focused optimization and decision systems for molecules, peptides, PROTACs, aptamers, and future modalities.
Research partnerships
Joint method development, benchmark design, evidence review, and project-specific computational studies.
06 / EasternDawn
Scientific intelligence that compounds across every program
EasternDawn builds drug discovery systems that preserve scientific context across retrieval, design, evaluation, and decision-making. Every new engine and agent expands the capability surface of the same composable platform.
- English name
- EasternDawn
- Chinese name
- 即白生科
- Legal entity
- 将东方之既白生物科技（深圳）有限公司