Fraglingo: Molecular Design via Attachment-Aware Autoregressive Fragment Generation

Thao Nguyen; Jeonghwan Kim; Zhenhailong Wang; Heng Ji

cs.AISep 11, 2026

Fraglingo: Molecular Design via Attachment-Aware Autoregressive Fragment Generation

Authors: Thao Nguyen, Jeonghwan Kim, Zhenhailong Wang, Heng Ji

Organizations: Siebel School of Computing and Data Science University of Illinois Urbana-Champaign Urbana, IL 61801, USA

Abstract

We introduce Fraglingo, an autoregressive molecular generator that constructs molecules step by step from chemically meaningful fragments connected through predefined attachment sites. At each generation step, Fraglingo jointly predicts which fragment to add and how it should attach by producing an attachment-aware fragment embedding and retrieving the nearest fragment through latent-space search. A wildcard-anchored readout represents both the growing molecule and candidate fragments relative to their attachment sites, enabling a single latent prediction to determine both fragment identity and attachment configuration. Because prediction operates in a continuous embedding space rather than over fixed fragment identifiers, larger fragment libraries can be introduced at inference time without retraining. This retrieval-based formulation provides a unified generation primitive for molecule generation, scaffold generation, scaffold decoration, and molecule optimization. Fraglingo also supports property-conditional generation, allowing desired molecular properties to guide the generation process. On controlled property-conditional benchmarks, Fraglingo achieves stronger joint property control than comparably trained baselines while maintaining competitive validity, uniqueness, and novelty. It also generalizes to fragment libraries up to four times larger than those used during training without retraining. Code is available at: https://anonymous.4open.science/r/FragLingo-3551.

Figures & tables

Term	Definition
Fragment	A chemically meaningful molecular substructure used as a generation unit.
Attachment site	A position on a fragment where a bond to another fragment can be formed, represented by a wildcard atom (*).
Active attachment site	The open attachment site of the growing molecule from which the next generation step proceeds.
Growing molecule	The partially assembled molecule at the current generation step.
Attachment-aware	A representation conditioned on the specific attachment site through which a fragment participates in assembly.
Wildcard-anchored readout	A graph readout that represents a molecule relative to a selected wildcard attachment site.

Table 1: Key terminology used in Fraglingo.

Figure 1: Attachment ambiguity in fragment-based molecular generation. The same pair of fragments can produce different molecular structures depending on which attachment sites are connected.

Figure 2: Overview of Fraglingo training and inference across four molecular design tasks. The architecture shown illustrates unconditional generation, with the GNN encoder and query projector abbreviated for clarity; detailed architectures are provided in Figure 8 .

Figure 3: (a) tSNE visualization of wildcard-anchored readouts. Each fragment contributes two readouts, one per wildcard; paired readouts are linked and shown with the same color. (b) fragments with the largest and smallest distances between their two wildcard-anchored readouts. Farthest pairs have distinct attachment environments, whereas closest pairs are nearly symmetric.

Model	Generation Quality				Joint Control				Per-Property Spearman
	Validity	Uniqueness	Novelty	Diversity	NJD	Joint@1 $\times$	Joint@2 $\times$	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
MolGPT	25.7	100.0	100.0	0.3099	1.7339	53.97	95.02	6.32	0.8072	0.9442	0.6659	0.8537	0.7340	0.8242	0.6421	0.7816
(Std)	(1.9)	(0)	(0)	(0.026)	(0.113)	(1.362)	(3.945)	(0.190)	(0.038)	(0.031)	(0.074)	(0.037)	(0.057)	(0.039)	(0.056)	(0.012)
MolRWKV	90.9	100.0	100.0	0.8261	1.6090	43.77	90.46	6.17	0.8606	0.4830	0.7070	0.6449	0.8839	0.8572	0.8232	0.7514
(Std)	(2.2)	(0)	(0)	(0.027)	(0.103)	(4.851)	(4.196)	(0.161)	(0.032)	(0.044)	(0.075)	(0.040)	(0.041)	(0.030)	(0.023)	(0.027)

Table 2: Property-conditional molecule generation benchmarking results. Each model is followed by a row reporting the standard deviation over 3 runs where available. Best results are shown in bold ; second-best results are underlined .

Figure 4: Examples from property-conditional molecule generation . Green marks properties close to the requested target, and yellow marks properties within the acceptable tolerance range.

Train	Vocab	Generation Quality				Joint Control				Per-Property Spearman
		Validity	Uniqueness	Novelty	Diversity	NJD	Joint@1 $\times$	Joint@2 $\times$	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
		( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
1K	100	97.8	98.7	100.0	0.8344	3.9288	12.24	42.58	3.99	0.5940	0.2953	0.2920	0.3903	0.5228	0.3101	0.4275	0.4045
	500	97.8	99.0	100.0	0.8631	4.2554	11.84	39.05	3.82	0.6623	0.1473	0.3138	0.4428	0.4705	0.3381	0.4717	0.4066
	1000	99.6	99.3	100.0	0.8748	4.1446	10.10	36.15	3.78	0.4501	0.1098	0.2547	0.4414	0.4598	0.3536	0.1643	0.3190
	2000	99.8	99.6	100.0	0.8775	4.3054	11.25	35.18	3.78	0.5069	0.1394	0.2646	0.3582	0.4166	0.2783	0.4005	0.3377

Table 3: Property-conditional molecule generation under different vocabulary sizes and training dataset sizes. Best results are shown in bold ; second-best results are underlined .

Inference Vocabulary	Generation Quality				Joint Control		Per-Property Spearman							Avg.	OOV Frag.
	Validity	Uniqueness	Novelty	Diversity	NJD	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
Training (500-frags)	100.0	99.9	100.0	0.8218	1.4928	6.56	0.8710	0.7746	0.6373	0.8296	0.8284	0.8190	0.7993	0.7942	0.00%
Inference (1000-frags)	100.0	100.0	100.0	0.8313	1.6028	6.40	0.8600	0.7496	0.6323	0.8476	0.8534	0.8280	0.7903	0.7945	29.11%
Inference (2000-frags)	100.0	100.0	100.0	0.8382	1.8628	6.16	0.8620	0.7286	0.5953	0.8226	0.8334	0.8150	0.7773	0.7763	48.76%
Training (1000-frags)	100.0	100.0	100.0	0.8313	1.2318	6.83	0.7740	0.8896	0.6043	0.8456	0.7354	0.7890	0.7523	0.7702	0.00%

Table 4: Inference-time vocabulary expansion. Models are trained with a fixed BFE fragment vocabulary but retrieve against a larger inference library at test time. OOV Frag.% is the fraction of selected fragments absent from the training vocabulary.

Appendix figures & tables11 assets

Supplementary material from the paper’s appendix.

Appendix

Configuration	Generation Quality				Joint Control				Per-Property Spearman
	Validity	Uniqueness	Novelty	Diversity	NJD	Joint@1 $\times$	Joint@2 $\times$	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
Full model (Fraglingo)	100.0	99.9	100.0	0.8218	1.4928	69.65	97.75	6.56	0.8710	0.7746	0.6373	0.8296	0.8284	0.8190	0.7993	0.7942
InfoNCE ( $\to$ MSE regression)	99.8	100.0	100.0	0.8238	4.2248	26.85	24.94	3.73	0.1010	0.3146	0.0033	0.3636	0.1674	0.1470	0.1453	0.1772
Classification loss	100.0	100.0	100.0	0.8228	1.7878	56.52	88.14	5.95	0.7570	0.6966	0.4823	0.7756	0.7434	0.7520	0.7193	0.7032
RVQ regularizer	100.0	100.0	100.0	0.8268	1.5268	67.75	95.76	6.29	0.7780	0.8536	0.6053	0.8046	0.7344	0.7980	0.7243	0.7562

Appendix

Table 5: Component ablation on the shared 10k-molecule, 500-fragment setting. Each row removes or replaces exactly one component relative to the full model (top row); rows are ordered by severity of the resulting drop in joint control.

Model	Generation Quality		Optimization Performance				Property Distribution Similarity
	Validity	Novelty	NJD	Tanimoto	Sim $\geq$ 0.6	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
	( $\uparrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
HierVAE	100.0%	100.0%	1.7717	0.6746	66.14%	6.17	0.821	0.723	0.727	0.719	0.817	0.809	0.719	0.762
(Std)	(0)	(0)	(0.011)	(0.014)	(2.6)	(0.006)	(0.010)	(0.012)	(0.009)	(0.010)	(0.014)	(0.004)	(0.004)	(0.011)
JT-VAE	100.0%	99.9%	2.6481	0.2940	4.12%	5.19	0.509	0.329	0.373	0.277	0.537	0.421	0.515	0.423
(Std)	(0)	(0)	(0.093)	(0.027)	(0.74)	(0.60)	(0.078)	(0.069)	(0.090)	(0.063)	(0.093)	(0.071)	(0.102)	(0.098)

Appendix

Table 6: Molecule optimization against fragment-based baselines on the shared 10k matched-pair dataset. Tanimoto and Sim

\geq

0.6 measure preservation of the starting molecule.

Figure 5: Qualitative molecule optimization examples. Green marks properties close to target, yellow marks properties within tolerance, and red marks properties outside the target range.

Train	Vocab	Generation	Optimization Performance				Property Distribution Similarity
		Validity	NJD	Tanimoto	Sim $\geq$ 0.6	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
		( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
10k	500	100.0%	1.6244	0.7952	95.72%	6.17	0.8390	0.7440	0.7130	0.7720	0.8690	0.8400	0.7880	0.7950
100k		100.0%	1.4031	0.7903	95.82%	6.34	0.8709	0.7712	0.7681	0.8327	0.8753	0.8593	0.8103	0.8269
1M		100.0%	1.2037	0.7853	94.62%	6.39	0.8617	0.7830	0.7981	0.8351	0.8941	0.8927	0.8791	0.8491
10k	1000	100.0%	1.6477	0.7896	94.72%	6.12	0.8414	0.6757	0.7280	0.7772	0.8656	0.8328	0.7924	0.7876

Appendix

Table 7: Ablation on training corpus size and vocabulary size for molecule optimization. Tanimoto and Sim

\geq

0.6 measure preservation of the starting molecule.

Figure 6: Qualitative scaffold generation examples. Fraglingo assembles scaffold-level BRICS fragments into valid core structures before side-chain decoration, producing diverse ring systems and attachment-ready topologies rather than complete decorated molecules.

Train	Vocab	Generation Quality				Joint Control				Per-Property Spearman
		Validity	Uniqueness	Novelty	Diversity	NJD	Joint@1 $\times$	Joint@2 $\times$	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
		( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
10k	50	100.0	96.8	88.2	0.8059	2.5943	41.67	84.64	5.68	0.6169	0.7711	0.4029	0.5954	0.6196	0.5987	0.6701	0.6107
	100	100.0	97.0	89.4	0.8262	2.5278	44.22	84.99	5.77	0.7397	0.7302	0.3994	0.6019	0.5859	0.6052	0.7786	0.6344
	200	100.0	99.0	95.2	0.8363	3.1045	32.05	74.45	5.28	0.7253	0.6606	0.2815	0.5352	0.6709	0.4859	0.5754	0.5621
	500	99.9	99.2	97.0	0.8397	3.9841	25.28	63.91	4.74	0.6526	0.4954	0.1639	0.4121	0.6040	0.2993	0.6391	0.4666

Appendix

Table 8: Scaffold generation under different vocabulary sizes and training dataset sizes.

Figure 7: Qualitative scaffold decoration examples. For each fixed scaffold, Fraglingo predicts branch fragments at the growable sites under the specified property targets. The tables report molecular-property values and signed errors before and after decoration, illustrating how site-conditioned retrieval can move decorated molecules toward the requested property profile while preserving the scaffold core.

Train	Vocab	Joint Control			Per-Property Spearman
Train	Vocab	NJD	Tanimoto	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
		( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
10k	500	1.08	0.3165	6.62	0.9126	0.7919	0.8381	0.8632	0.8996	0.8538	0.8032	0.8518
	1000	0.95	0.3338	6.75	0.9115	0.8711	0.8852	0.9194	0.8839	0.9225	0.8891	0.8975
	2000	1.00	0.3206	6.70	0.9030	0.8290	0.8287	0.8965	0.8770	0.9069	0.8870	0.8755
100k	500	0.80	0.3349	6.75	0.9412	0.8488	0.9086	0.9469	0.8826	0.9324	0.8958	0.9080

Appendix

Table 9: Scaffold decoration under different vocabulary and corpus sizes. Metrics are computed on decorated molecules with the scaffold held fixed.

Figure 8: Detailed architecture of Fraglingo’s next-fragment generation module.

Model	Dim.	Layers	Heads	Params
Small	128	4	4	1.2M
Baseline	256	6	8	6.4M
Large	512	8	8	32.9M

Appendix

Table 10: Model configurations and parameter counts.

Model size	Generation Quality				Joint Control				Per-Property Spearman
	Validity	Uniqueness	Novelty	Diversity	NJD	Joint@1 $\times$	Joint@2 $\times$	Partial (/7)	logP	MW	QED	TPSA	HBD	HBA	RotBonds	Avg.
	( $\uparrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\downarrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )	( $\uparrow$ )
1.2M	100%	100%	100%	0.843	2.011	38.9%	84.9%	5.80	0.823	0.611	0.568	0.756	0.821	0.776	0.756	0.730
6.4M	100%	100%	100%	0.8444	1.9808	40.88%	85.90%	5.80	0.8373	0.6423	0.5632	0.7605	0.8161	0.7800	0.7760	0.7393
32.9M	100%	100%	100%	0.8443	2.017	39.05%	83.70%	5.73	0.793	0.621	0.603	0.754	0.825	0.782	0.750	0.733
6.4M ∗	100%	99.9%	100%	0.8218	1.4928	69.65%	97.75%	6.56	0.8710	0.7746	0.6373	0.8296	0.8284	0.8190	0.7993	0.7942

Appendix

Table 11: Effect of model size on property-conditional molecule generation (10k corpus, 500 fragments). Scaling parameters improves joint control at saturated generation quality. The 6.4M ∗ row applies BPE on BRICS fragments, called BFE (BRICS-based Fragment Enumeration) in MolLingo, coarsening the vocabulary.

Explore similar work

Oct 11, 2026

FragFlex: Fragment-Based Flexible Molecular Generation for Exploring Vast Chemical Space

Drug discovery requires efficient exploration of an astronomically large chemical space. Molecular graph diffusion models offer a promising approach to this task, but many require graph size to be fixed before generation. Although atom-level insertion and deletion can enable flexible-size generation, jointly modeling molecular size, atom types, and bonds while maintaining chemical validity remains challenging. We introduce FragFlex, a flexible-length discrete diffusion model that performs node insertion and deletion at the fragment level. During denoising, FragFlex jointly learns the size, fragment identities, and inter-fragment connectivity of the fragment graph. A neural attachment-site predictor and valence-aware assembler then convert the generated fragment graph into an atom-level molecule. In large-scale experiments generating 10 million molecules, FragFlex achieves near-perfect chemical validity while maintaining high uniqueness, demonstrating a strong balance between the two.

Sihyun Park, David Ryan Koes

Carnegie Mellon University · University of Pittsburgh

Molecular Graph Generation Molecular Generation Diffusion Models Drug Discovery

Ask PDF

Nov 10, 2024

MolMiner: Toward Controllable, 3D-Aware, Fragment-Based Molecular Design

We introduce MolMiner, a fragment-based, geometry-aware, and order-agnostic autoregressive model for molecular design. MolMiner supports high-dimensional conditional control over twelve physicochemical and structural properties from partial specifications, constructs molecules via symmetry-aware fragment attachments, and conditions each generation step on force-field-relaxed three-dimensional geometry of the partial structure. Conditional control emerges without auxiliary property losses. On targeted property windows, conditioning lifts hit rates by up to 5.25x over unconditional generation and 3.5x over the training distribution itself -- overriding the model's intrinsic biases -- at the cost of a small reduction in unconditional distributional fidelity. MolMiner unifies dynamic geometry, symmetry handling, order-agnostic generation, and scalable multi-property conditioning within a single framework.

Raul Ortega-Ochoa, Tejs Vegge, Jes Frellsen

Department of Energy Conversion and Storage, Technical University of Denmark, DK-2800 Kgs. Lyngby, Denmark · CAPeX Pioneer Center for Accelerating P2X Materials Discovery · Department of Applied Mathematics and Computer Science, Technical University of Denmark

3D Molecular Generation Autoregressive Generation Molecular Generation Conditional Generation

Ask PDF

Jul 10, 2026

Autoregressive latent diffusion for 3D molecule generation

Three-dimensional (3D) molecule generation has been dominated by diffusion models, which achieve strong generation quality but typically require the molecular size to be specified a priori. Recent autoregressive approaches have substantially narrowed the performance gap while naturally supporting variable-length generation and conditioning on partial molecular context. However, balancing unconditional and context-conditioned generation remains challenging. We introduce KRONOS, a latent autoregressive diffusion framework that generates molecules in the latent space of a pre-trained autoencoder, jointly modeling molecular graph topology and geometry, while retaining the flexibility of autoregressive generation. We further introduce a mixed training strategy inspired by Fill-in-the Middle (FIM) paradigm, enabling both unconditional and fragment-conditioned molecular generation within a single left-to-right autoregressive model. Experiments on QM9 and GEOM-Drugs demonstrate that KRONOS achieves leading unconditional generation performance among autoregressive methods, while remaining competitive with diffusion models. Moreover, fragment-conditioned generation is achieved with negligible impact on unconditional generation performance, demonstrating that both generation paradigms can be supported within a single architecture.

Federico Ottomano, Gaopeng Ren, Yingzhen Li +2

Department of Chemistry Imperial College London · Department of Computing Imperial College London

Latent Diffusion Models Molecular Graph Generation 3D Molecular Generation Autoregressive Generation Molecular Generation

Ask PDF