MADDIE: Multi-Agent Drug Discovery Engine

Ye Liu; Reza Averly; Avery Meyer; Frazier N. Baker; Ruoxi Gao; Daniel Adu-Ampratwum; Xia Ning

cs.AIOct 12, 2026

MADDIE: Multi-Agent Drug Discovery Engine

Authors: Ye Liu, Reza Averly, Avery Meyer, Frazier N. Baker, Ruoxi Gao, Daniel Adu-Ampratwum, Xia Ning

Organizations: Biomedical Informatics, The Ohio State University. · Computer Science and Engineering, The Ohio State University. · Medicinal Chemistry and Pharmacognosy, The Ohio State University. · Translational Data Analytics Institute, The Ohio State University.

Abstract

Drug discovery remains a resource-intensive and time-consuming process, involving billions of dollars and a decade of development to advance a drug to market. While many in silico tools have accelerated this process, operating them still demands substantial time, effort, and expertise from medicinal chemists. Autonomous systems built on Large Language Models (LLMs) have emerged as a promising solution, yet current designs limit their performance. Existing single-agent systems scale poorly to long-horizon, multi-step, and realistic tasks, whereas multi-agent systems typically execute predefined workflows that cannot capture the iterative, non-linear nature of a real drug discovery campaign. Here, we present MADDIE, a multi-agent drug discovery engine. It comprises a central agent, Orchestrator, and four specialized worker agents: HitIdentifier, LeadOptimizer, Synthesizer, and Evaluator, each representing a critical step in a drug discovery campaign. Under Orchestrator's orchestration and coordination, these worker agents are composed into flexible workflows, enabling MADDIE to adapt its strategy and navigate the drug discovery pipeline end-to-end. Across 50 protein targets, MADDIE produces molecules satisfying key property requirements and their synthesis routes, outperforming existing agentic systems, generative models, and general-purpose LLMs by a wide margin. Our analysis reveals how these agents collaborate and that MADDIE can employ different strategies across targets to achieve the campaign's goal, while still being efficient in time and tokens. Together, they demonstrate MADDIE as a low-cost, effective, and efficient autonomous engine for accelerating drug discovery. Code and dataset is available via https://github.com/ninglab/MADDIE.

Figures & tables

Figure 1: $\mathop{\textsc{MADDIE}}\limits$ architecture, consisting of $\mathop{\textsc{Orchestrator}}\limits$ , workspace and four worker agents: $\mathop{\textsc{HitIdentifier}}\limits$ , $\mathop{\textsc{LeadOptimizer}}\limits$ , $\mathop{\textsc{Synthesizer}}\limits$ and $\mathop{\textsc{Evaluator}}\limits$ .

Agents for Drug Discovery
Method	$\mathop{\mathtt{Vina}}\limits$	$\mathop{\mathtt{QED}}\limits$	$\mathop{\mathtt{SAS}}\limits$	$\mathop{\mathtt{Mut}}\limits$	$\mathop{\mathtt{CTx}}\limits$	$\mathop{\mathtt{hERG}}\limits$	$\mathop{\mathtt{Lly}}\limits$	$\mathop{\mathtt{Syn}}\limits$
Method	$<$ -6.53	$>$ 0.36	$<$ 5.35	$<$ 0.51	$<$ 0.71	$<$ 0.96	True	True
$\mathop{\textsc{MADDIE}}\limits$ (ours)	✓	✓	✓	✓	✓	✓	✓	✓
$\mathop{\textsc{LIDDIA}}\limits$ ( Averly et al., 2025 )	✓	✓	✓	–	–	–	–	–
Prompt-to-Pill ( Vichentijevikj et al., 2026 )	✓	✓	–	✓	–	✓	–	–
AgentD ( Ock et al., 2026 )	–	✓	–	–	–	–	–	–

Table 1: The thresholds used in the experiments and whether each method can measure the corresponding properties and synthesizability during generation (✓ indicates it can). The thresholds are calculated using marketed FDA-approved drugs in our dataset (Appendix 9 ).

Agents for Drug Discovery
Method	$\mathop{\textbf{$ N_{tout} $}}\limits$	$\mathop{\textbf{$ N_{out} $}}\limits$	$\mathop{\textbf{$ N_{sat} $}}\limits$	% $\mathop{\texttt{Pass}}\limits$	% $\mathop{\mathtt{Vina}}\limits$	% $\mathop{\mathtt{QED}}\limits$	% $\mathop{\mathtt{SAS}}\limits$	% $\mathop{\mathtt{Mut}}\limits$	% $\mathop{\mathtt{CTx}}\limits$	% $\mathop{\mathtt{hERG}}\limits$	% $\mathop{\mathtt{Lly}}\limits$	% $\mathop{\mathtt{Syn}}\limits$
$\mathop{\textsc{MADDIE}}\limits$	50	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0
$\mathop{\textsc{LIDDIA}}\limits$	49	36.0	28.0	80.0	83.1	96.1	100.0	100.0	100.0	100.0	100.0	100.0
Prompt-to-Pill	31	14.2	2.9	20.9	55.0	88.5	100.0	41.0	100.0	98.1	100.0	100.0
AgentD	35	44.2	44.2	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0
Generative Models for Drug Design

Table 2: Main experimental results of

\mathop{\textsc{MADDIE}}\limits

against baselines.

Method	$\mathop{\textbf{$ N_{tout} $}}\limits$	$\mathop{\textbf{$ N_{out} $}}\limits$	$\mathop{\textbf{$ N_{sat} $}}\limits$	% $\mathop{\texttt{Pass}}\limits$	% $\mathop{\mathtt{Syn}}\limits$	$\mathop{\textbf{$ N_{itm} $}}\limits$
$\mathop{\textsc{MADDIE}}\limits$	50	100.0	100.0	100.0	100.0	1,043.8
w/o $\mathop{\textsc{Synthesizer}}\limits$	50	100.0	40.2	40.2	40.2	1,554.8
w/o $\mathop{\textsc{LeadOptimizer}}\limits$	49	100.0	100.0	100.0	100.0	1,327.7
w/o $\mathop{\textsc{Orchestrator}}\limits$	50	100.0	100.0	100.0	100.0	1,743.8

Table 3: Ablation study.

Figure 2: Transition probabilities among worker agents during a campaign. Red color represents the transition with the highest probability out of each worker agent.

Panel of Figure 3 — Figure 3: Flow of molecules from the worker agent at which they originate to the campaign output. Each subfigure traces the molecules originating from one worker agent; each edge gives the number of those molecules subsequently passed to another worker agent or reaching the output, averaged over the 50 targets.

Panel of Figure 4 — Figure 4: Trajectories of $\mathop{\textsc{MADDIE}}\limits$ over six representative targets: (a) ABCD1_ALD, (b) BCHE, (c) ADAM10, (d) ADRB1, (e) IDH2, and (f) PAK4. Rows represent the agents in $\mathop{\textsc{MADDIE}}\limits$ and columns represent one delegation or step during the campaign. The number on the arrow denotes the number of molecules passed to the next agent. The number on the box denotes the number of molecules stored in each repository after the delegation. The arrow width (worker agents) and box size ( $\mathop{\textsc{Orchestrator}}\limits$ ) are proportional to the number of molecules.

Panel of Figure 5 — Figure 5: PSEN1 case study.

$\mathop{\textbf{$ N_{itm} $}}\limits$	Targets	Share (%)
$<750$	10	20.0
$750$ – $1{,}000$	18	36.0
$1{,}000$ – $1{,}250$	9	18.0
$1{,}250$ – $1{,}500$	8	16.0
$\geq 1{,}500$	5	10.0

Table 4: Number of targets grouped by the number of intermediate molecules explored.

Metric	Cost		Action			Delegation per agent
Metric	Time (h)	Out. token	Total	Process	Delegate	HI	LO	EV	SY
Average	3.3	21,125	24.9	14.9	10.0	2.2	1.0	5.1	1.7
Shortest (ADAMTS5)	1.2	7,508	9	5	4	1	0	2	1
Longest (HDAC8)	11.6	30,448	44	38	6	1	1	3	1

Table 5: Computational cost of

\mathop{\textsc{MADDIE}}\limits

over 50 targets.

Worker agent	Input	Output	Tools
$\mathop{\textsc{HitIdentifier}}\limits$	Target pocket $P$ .	Candidate molecules.	conDitar ( Gao et al., 2026 ) for SBDD; DiffSMol ( Chen et al., 2025 ) for LBDD; AutoDock ( Trott and Olson, 2010a ) for predicting binding affinity.
$\mathop{\textsc{Evaluator}}\limits$	Candidate molecules.	Property values.	UniDock ( Yu et al., 2023 ) , QED ( Bickerton et al., 2012 ) , SAScore ( Ertl and Schuffenhauer, 2009 ) , ADMET-AI ( Swanson et al., 2024 ) and Lilly filter ( Bruns and Watson, 2012 ) .
$\mathop{\textsc{LeadOptimizer}}\limits$	Candidate molecules that fail one or more requirements.	Optimized candidate molecules.	GeLLMO-C ( Dey et al., 2025 ) ; MolOpt algorithms ( Gao et al., 2022 ) ; RDKit and $\mathop{\textsc{Evaluator}}\limits$ -backed oracles for the target properties.
$\mathop{\textsc{Synthesizer}}\limits$	Candidate molecules.	Synthesis routes or synthesizable analogs.	MEEA ∗ ( Zhao et al., 2024 ) and PrexSyn ( Luo and Coley, 2026 ) .

Table 6: Artifacts required and produced by each worker agent, along with their respective tools.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Gene name	PDB ID	Ligand ID	Disease category	Source
				$\mathop{\textsc{LIDDIA}}\limits$	CD2020
ABCC8	6PZ9	BJX	D	–	–
ABCD1	7VX8	ATP	N	–	$\checkmark$
ABO	5BXC	DA8	CV, I	–	$\checkmark$
ACE	3L3N	LSW	CV	–	$\checkmark$
ACHE	7D9O	H0L	A, N	–	–

Appendix

Table A1 : Protein targets in the

\mathop{\textsc{MADDIE}}\limits

dataset.

Target	Associated FDA-approved drugs
ABCC8	Glimepiride, Glipizide, Glyburide, Nateglinide, Repaglinide
ACE	Benazepril, Captopril, Enalapril, Enalaprilat, Fosinopril, Lisinopril, Moexipril, Perindopril, Quinapril, Ramipril, Trandolapril
ACHE	Benzgalantamine, Donepezil, Echothiophate, Galantamine, Neostigmine, Pralidoxime, Pyridostigmine, Rivastigmine
ADRB1	Acebutolol, Atenolol, Betaxolol, Bisoprolol, Carteolol, Carvedilol, Dobutamine, Dopamine, Epinephrine, Esmolol, Isoproterenol, Labetalol, Landiolol, Levobunolol, Metoprolol, Nadolol, Nebivolol, Norepinephrine, Pindolol, Propafenone, Propranolol, Sotalol, Timolol
AKT1	Capivasertib
BCHE	Rivastigmine

Appendix

Table A2 : FDA-approved drug associations for targets in the

\mathop{\textsc{MADDIE}}\limits

dataset.

Pass rate	Requirements	FDA-approved drugs
% $\mathop{\mathtt{Vina}}\limits$	$<-6.53$	81.2
% $\mathop{\mathtt{QED}}\limits$	$>0.36$	88.0
% $\mathop{\mathtt{SAS}}\limits$	$<5.35$	96.9
% $\mathop{\mathtt{Mut}}\limits$	$<0.51$	91.1
% $\mathop{\mathtt{CTx}}\limits$	$<0.71$	89.0
% $\mathop{\mathtt{hERG}}\limits$	$<0.96$	81.7

Appendix

Table A3 : Property requirements and pass rates for our dataset of FDA-approved drugs

Agents for Drug Discovery
Method	$\mathop{\mathtt{Vina}}\limits$ $\downarrow$	$\mathop{\mathtt{QED}}\limits$ $\uparrow$	$\mathop{\mathtt{SAS}}\limits$ $\downarrow$	$\mathop{\mathtt{Mut}}\limits$ $\downarrow$	$\mathop{\mathtt{CTx}}\limits$ $\downarrow$	$\mathop{\mathtt{hERG}}\limits$ $\downarrow$	$\mathop{\mathtt{Div}}\limits$ $\uparrow$	Avg. Rank $\downarrow$
Method	-6.530	0.360	5.350	0.510	0.710	0.960	–	–
$\mathop{\textsc{MADDIE}}\limits$	-7.897	0.646	3.563	0.217	0.224	0.535	0.864	6.3
$\mathop{\textsc{LIDDIA}}\limits$	-7.780	0.712	2.515	0.226	0.108	0.358	0.811	4.1
Prompt-to-Pill	-6.834	0.570	3.378	0.541	0.221	0.356	0.739	8.3
AgentD	-8.223	0.702	2.861	0.221	0.217	0.542	0.693	6.1

Appendix

Table A4 : Average property values over output molecules from all targets

Panel of Figure 16 — Figure A1 : Transition trajectories of $\mathop{\textsc{MADDIE}}\limits$ on two representative targets.

Figure A2 :

\mathop{\textsc{MADDIE}}\limits

’s generated molecules on PSEN1 exceeding either the reference ligand or Nirogacestat on all in silico properties defined in Section 3 .

Panel of Figure 18 — Figure A3 : ACHE case study

Panel of Figure 19 — Figure A4 : BCHE case study

Explore similar work

Mozi: Governed Autonomy for Drug Discovery LLM Agents
Mar 4, 2026He Cao, Siyu Liu, Fan Zhang +7AI Agents for Scientific Discovery LLM Agent Orchestration
DrugSAGE:Self-evolving Agent Experience for Efficient State-of-the-Art Drug Discovery
May 14, 2026Yikun Zhang, Xiwei Cheng, Tianyu Liu +2AI Agents for Scientific Discovery AI-Assisted Scientific Research
1Northeastern University · 2Broad Institute of MIT and Harvard · 3Yale University +1
From Benchmark to Bench: Can Agents Survive Real-World Drug Discovery?
Oct 5, 2026Pierre Llompart, Levent Guner, Helen Lai +2AI Agents for Scientific Discovery LLM Agent Orchestration
AMD Silo AI, Stockholm, Sweden · Molecular AI, Discovery Sciences, R&D, AstraZeneca, Cambridge, UK · Molecular AI, Discovery Sciences, R&D, AstraZeneca, Gothenburg, Sweden +1

Mar 4, 2026J/K move · Enter open · S save

Mozi: Governed Autonomy for Drug Discovery LLM Agents

He Cao, Siyu Liu, Fan Zhang, Zijing Liu+6

AI Agents for Scientific Discovery LLM Agent Orchestration Long-Horizon LLM Agents AI Agent Governance Drug Discovery

Open paper AskPDF arXiv

Tool-augmented large language model (LLM) agents promise to unify scientific reasoning with computation, yet their deployment in high-stakes domains like drug discovery is bottlenecked by two critical barriers: unconstrained tool-use governance and poor long-horizon reliability. In dependency-heavy pharmaceutical pipelines, autonomous agents often drift into irreproducible trajectories, where early-stage hallucinations multiplicatively compound into downstream failures. To overcome this, we present Mozi, a dual-layer architecture that bridges the flexibility of generative AI with the deterministic rigor of computational biology. Layer A (Control Plane) establishes a governed supervisor--worker hierarchy that enforces role-based tool isolation, limits execution to constrained action spaces, and drives reflection-based replanning. Layer B (Workflow Plane) operationalizes canonical drug discovery stages -- from Target Identification to Lead Optimization -- as stateful, composable skill graphs. This layer integrates strict data contracts and strategic human-in-the-loop (HITL) checkpoints to safeguard scientific validity at high-uncertainty decision boundaries. Operating on the design principle of ``free-form reasoning for safe tasks, structured execution for long-horizon pipelines,'' Mozi provides built-in robustness mechanisms and trace-level audibility to completely mitigate error accumulation. We evaluate Mozi on PharmaBench, a curated benchmark for biomedical agents, demonstrating superior orchestration accuracy over existing baselines. Furthermore, through end-to-end therapeutic case studies, we demonstrate Mozi's ability to navigate massive chemical spaces, enforce stringent toxicity filters, and generate highly competitive in silico candidates, effectively transforming the LLM from a fragile conversationalist into a reliable, governed co-scientist.

Abstract

Explore similar work

Mozi: Governed Autonomy for Drug Discovery LLM Agents

DrugSAGE:Self-evolving Agent Experience for Efficient State-of-the-Art Drug Discovery

From Benchmark to Bench: Can Agents Survive Real-World Drug Discovery?