Coverage Before Control: Route-Instruction Grounding and Steering for Controllable Retrosynthesis

Xuemin Chen; Xiaozhuang Song; Xinjian Zhao; Yaoyao Xu; Tianshu Yu

cs.AISep 30, 2026

Coverage Before Control: Route-Instruction Grounding and Steering for Controllable Retrosynthesis

Authors: Xuemin Chen, Xiaozhuang Song, Xinjian Zhao, Yaoyao Xu, Tianshu Yu

Organizations: The Chinese University of Hong Kong, Shenzhen

Abstract

Single-step retrosynthesis models are commonly evaluated by their ability to recover recorded reactions. In practice, chemists may need to choose among several precursor sets for the same product, for example to preserve a particular motif. Recovering a recorded answer alone does not establish this ability to follow a preference. Satisfying such requests requires both coverage of relevant alternatives and control over which alternatives are favored. We introduce Route-Instruction Grounding and Steering (RIGS), a two-stage framework for instruction-conditioned retrosynthesis. Stage A trains a language projector, teaching it which alternatives an instruction favors or discourages. Stage B uses the projector learned in Stage A to steer a frozen generative model through lightweight residual adapters. We construct nested one-to-many training supports by pairing each product with increasing numbers of candidate precursor sets. Extensive experiments demonstrate that broader support helps the model generate a wider range of alternatives, and RIGS can learn to guide generation according to instructions. The relationship between coverage and control is consistent across model scales but non-monotone.

Figures & tables

Figure 2: RIGS-GraphDiT: grounding and steering. Stage A grounds natural-language guidance against preferred and avoided routes. Stage B transfers this grounding into a residual instruction pathway that steers probability within the frozen generator’s learned support.

Figure 2

Figure 3: Candidate coverage and instruction control across training supports. GraphDiT-8M (top) and 65M (bottom) on the same Top100 test inventory, using equal-axis, case-balanced means. Left panels show Reference: (a,e) the rate of sampling at least one candidate from this inventory and (b,f) the number of distinct valid outputs per group, with raw-draw budget

B

on the vertical axis. Right panels compare Matched with Shuffled instructions at

B=100

: (c,g) increases in preferred-set hit rate,

\Delta\mathrm{P}@k

, and (d,h) reductions in avoided-set hit rate,

-\Delta\mathrm{N}@k

, for

k=1,10

. Positive values indicate better instruction alignment. Control markers show means over three Stage-B training seeds; shaded bands span their minimum and maximum. Each seed uses its validation-selected checkpoint and guidance scale; Reference is fixed for each support. Result summaries and seed analyses can be found in Appendices B and C.4 , respectively.

	Reference	Absolute P@1 (%)		Matched–Shuffled (pp)				Matched–Reference (pp)
Support	Pooled recovery	Reference	Matched	$\Delta\mathrm{P}@1$	$\Delta\mathrm{P}@10$	$-\Delta\mathrm{N}@1$	$-\Delta\mathrm{N}@10$	$\Delta\mathrm{P}@1$	$\Delta\mathrm{P}@10$	$-\Delta\mathrm{N}@1$	$-\Delta\mathrm{N}@10$
GraphDiT-8M
USPTO-MR	2,348	8.48	8.98	+0.08	+2.03	+0.74	+1.37	+0.51	+1.91	+0.86	+2.54
Top15	12,880	30.16	34.10	+4.10	+1.37	+3.83	+4.65	+3.95	+0.27	+4.06	+5.16
Top30	16,922	30.20	40.20	+12.89	+10.70	+14.06	+18.52	+10.00	-0.66	+18.24	+27.73
Top45	17,370	29.84	40.63	+14.18	+9.53	+14.06	+18.79	+10.78	+2.62	+15.20	+24.92

Table 1: Candidate coverage and instruction following at validation-selected settings. Results use

B=100

per group and Stage-B seed 42. Pooled recovery counts distinct product–candidate pairs recovered by Reference from the Top100 inventory across all instruction groups. Effects are in percentage points; positive values indicate improvement. Bold marks maxima in the effect columns within each model size. Absolute rates and Matched–Shuffled intervals appear in Tables 8 and 7 , respectively. Matched–Reference intervals for P@1, P@10, and N@10 appear in Table 9 . Three-seed summaries can be found in Appendix C.4 .

Table 5

			Matched–Shuffled (pp)		Matched–Reference (pp)
Reaction data	Guidance	Backbone	$\Delta$ P@1 $\uparrow$	$-\Delta$ N@10 $\uparrow$	$\Delta$ P@1 $\uparrow$	$-\Delta$ N@10 $\uparrow$
Pistachio	RDKit5	65M	+1.52	+7.19	+0.98	+13.59
USPTO	Structure	8M	+3.55	+3.43	+5.54	+10.57
		65M	+8.09	+18.44	+14.74	+35.79
USPTO	Metric4	8M	+5.37	+7.41	+4.07	+8.04
		65M	+15.78	+21.92	+13.44	+27.52

Table 4: Control across datasets and guidance suites. We evaluate separately trained Top30 models for each dataset and guidance suite. Positive values indicate better instruction alignment. Cohort and aggregation details for Pistachio, Structure, and Metric4 can be found in Appendices G.1 , F.1 , and G.2 , respectively.

Appendix figures & tables34 assets

Supplementary material from the paper’s appendix.

Appendix

Guided model ID	Stage-B epoch	$\alpha_{\mathrm{eval}}$
USPTO–Guide(RDKit5)–8M–USPTO-MR	14	.50
USPTO–Guide(RDKit5)–8M–Top15	6	.20
USPTO–Guide(RDKit5)–8M–Top30	12	.50
USPTO–Guide(RDKit5)–8M–Top45	14	.50
USPTO–Guide(RDKit5)–8M–Top60	6	.25
USPTO–Guide(RDKit5)–65M–USPTO-MR	14	.50

Appendix

Table 5: Frozen coordinates for the ten USPTO–Guide(RDKit5)–8M–[Support] and USPTO–Guide(RDKit5)–65M–[Support] models in the Common Top100 study. We selected all coordinates on validation before opening the common test; every row uses Reference, Matched, and Shuffled arms.

GraphDiT-8M
Support	Hit@1	Hit@100	Pooled	Unique valid /group	Validity	Duplicate rate
USPTO-MR	6.17	59.30	2,348	37.475	63.49	26.01
Top15	21.91	80.82	12,880	44.044	68.38	24.34
Top30	18.91	82.89	16,922	53.074	68.09	15.01
Top45	16.56	83.24	17,370	55.828	67.36	11.53
Top60	14.77	82.77	17,378	57.067	66.14	9.07

Appendix

Table 6: Reference coverage at the endpoints of the Common Top100 budget curves. Hit, validity and duplicate rate are percentages; pooled recovery counts distinct case–candidate pairs. Rates and mean counts use equal-axis, case-balanced aggregation at both capacities. All columns except Hit@1 use

B=100

, without backfilling.

GraphDiT-8M
Support	$\Delta$ P@1	$\Delta$ P@10	$\Delta$ N@1	$\Delta$ N@10
USPTO-MR	$+0.08$ $[-0.66,+0.82]$	$+2.03$ $[+1.05,+3.05]$	$-0.74$ $[-1.48,0.00]$	$-1.37$ $[-2.42,-0.35]$
Top15	$+4.10$ $[+2.85,+5.31]$	$+1.37$ $[+0.55,+2.19]$	$-3.83$ $[-4.92,-2.70]$	$-4.65$ $[-5.70,-3.59]$
Top30	$+12.89$ $[+11.02,+14.77]$	$+10.70$ $[+9.22,+12.19]$	$-14.06$ $[-15.70,-12.42]$	$-18.52$ $[-20.27,-16.80]$
Top45	$+14.18$ $[+12.46,+15.94]$	$+9.53$ $[+8.12,+10.94]$	$-14.06$ $[-15.70,-12.42]$	$-18.79$ $[-20.47,-17.11]$
Top60	$+5.63$ $[+4.06,+7.15]$	$+4.88$ $[+3.67,+6.13]$	$-5.94$ $[-7.46,-4.45]$	$-8.36$ $[-9.73,-6.95]$

Appendix

Table 7: Matched–Shuffled effects at returned ranks 1 and 10 on Common Top100. Entries are percentage-point estimates with 95% paired case-cluster bootstrap intervals, using equal-axis, case-balanced aggregation. Positive

\Delta

P and negative

\Delta

N are favorable. The registered primary coordinates use Stage-B seed 42.

GraphDiT-8M
Support	Arm	P@1	P@10	N@1	N@10	Validity
USPTO-MR	Reference	8.477	25.547	8.555	25.391	63.488
	Matched	8.984	27.461	7.695	22.852	60.816
	Shuffled	8.906	25.430	8.438	24.219	60.913
Top15	Reference	30.156	58.789	30.117	58.242	68.382
	Matched	34.102	59.062	26.055	53.086	67.084

Appendix

Table 8: Key absolute-arm Common Top100 results (%) at the coordinates in Table 5 . Each support lists all three conditions at the same raw-draw budget

B=100

GraphDiT-8M
Support	$\Delta$ P@1	$\Delta$ P@10	$-\Delta$ N@10
USPTO-MR	$+0.51$ $[-0.23,+1.25]$	$+1.91$ $[+0.86,+2.97]$	$+2.54$ $[+1.45,+3.63]$
Top15	$+3.95$ $[+2.66,+5.16]$	$+0.27$ $[-0.55,+1.09]$	$+5.16$ $[+4.14,+6.17]$
Top30	$+10.00$ $[+8.09,+11.87]$	$-0.66$ $[-2.03,+0.74]$	$+27.73$ $[+25.90,+29.53]$
Top45	$+10.78$ $[+9.06,+12.50]$	$+2.62$ $[+1.33,+3.91]$	$+24.92$ $[+23.24,+26.72]$
Top60	$+3.83$ $[+2.27,+5.39]$	$+1.17$ $[-0.08,+2.42]$	$+10.08$ $[+8.71,+11.48]$

Appendix

Table 9: Returned Top-

k

changes relative to unguided generation. Cells give Matched–Reference effects in percentage points with 95% paired case-bootstrap intervals. We reverse the sign of avoided-set recovery; larger values are favorable in every column. Both capacities show early-rank preferred gains at Top15/30/45, while higher-rank effects vary.

Figure 4: Common-update-ceiling support-scaling sensitivity. Points are equal-axis, case-balanced Matched–Shuffled effects on the frozen Common Top100 test, shown as probability differences (

0.10=10

percentage points). Error bars are 95% shared case-cluster bootstrap intervals. The right panel plots

-\Delta\mathrm{N}@1

so that larger values are favorable. Top15/30/45 use the common update ceilings above; Top60 reuses its primary checkpoint.

8M
Support	Selected B (coord.)	$\alpha$	$\Delta$ P@1	$\Delta$ N@1	$\Delta V$
Top15	58,730 (14)	$.15$	$+2.30$ $[+1.17,+3.44]$	$-3.63$ $[-4.73,-2.62]$	$+0.07$ $[-0.06,+0.20]$
Top30	33,560 (8)	$.50$	$+11.21$ $[+9.61,+12.85]$	$-10.94$ $[-12.54,-9.34]$	$-0.05$ $[-0.30,+0.20]$
Top45	50,340 (12)	$.25$	$+6.48$ $[+4.96,+8.01]$	$-5.82$ $[-7.15,-4.45]$	$+0.07$ $[-0.11,+0.26]$
Top60	reused	$.25$	$+5.62$ $[+4.06,+7.15]$	$-5.94$ $[-7.46,-4.45]$	$-0.00$ $[-0.21,+0.20]$
65M

Appendix

Table 10: Common-update-ceiling deployment coordinates and effects. Selected B gives optimizer updates (Top60-equivalent epoch index); reused denotes the primary Top60 coordinates in Table 5 . Effects are Matched minus Shuffled in percentage points [95% shared case-cluster CI]; larger

\Delta

P@1 and smaller

\Delta

N@1 are favorable.

V

denotes raw-draw validity.

Reverse $K$	Cov.	Acc.	Cov.	Acc.	Cov.	Acc.	Cov.	Acc.
	Reference		Guidance		Shuffled		$\Delta$ (pp)
1	63.00	63.00	71.00	71.00	57.50	57.50	$+8.00$	$+8.00$
3	76.50	59.50	75.00	62.83	73.50	56.83	$-1.50$	$+3.33$
5	79.00	57.90	76.50	57.90	76.00	52.60	$-2.50$	$0.00$
10	80.00	51.25	80.00	47.85	77.00	42.90	$0.00$	$-3.40$

Appendix

Table 11: Relaxed Forward Top-1 round-trip results (%) for Top30 65M on 200 frozen test groups. Reverse

K

is the candidate-rank cutoff. Coverage is the fraction of groups with at least one round-trip hit; Accuracy uses the fixed

200K

candidate-slot denominator, counting missing or invalid ranks as failures. Guidance denotes matched guidance, and Shuffled is the shuffled-guidance control.

\Delta

is Guidance minus Reference in percentage points. Bold marks the highest point estimate for each metric and row. All results include the reference shortcut in Eq. ( 7 ).

Figure 5: Guidance-strength trade-off at fixed checkpoints. GraphDiT-65M on each support’s original validation cohort at epoch 14. P@1 gain and raw-validity loss are relative to

\alpha=0

. Stars mark final selected scales evaluated at this checkpoint; deployed checkpoints may differ.

Support	$\Delta$ P@1	$\Delta$ P@10	$\Delta$ N@1	$\Delta$ N@10	$\Delta V$
USPTO-MR	+0.04	+0.63	-0.16	-1.25	-0.04
Top15	+4.73	+2.27	-4.69	-5.00	-0.05
Top30	+4.80	+3.28	-4.22	-8.67	-0.08
Top45	+6.60	+4.41	-6.37	-9.61	+0.01
Top60	+5.63	+4.88	-5.94	-8.36	+0.00

Appendix

Table 12: Matched-scale 8M sensitivity at

\alpha_{\mathrm{eval}}=.25

. Values are equal-axis, case-balanced Matched–Shuffled effects in percentage points. Positive

\Delta\mathrm{P}

and negative

\Delta\mathrm{N}

indicate better instruction alignment;

\Delta V

is the Matched–Shuffled trajectory-validity difference.

Figure 6: Fixed-checkpoint scale response across three Stage-B seeds. GraphDiT-8M Top45 and Top60: (a) Reference-minus-Matched trajectory-validity loss; (b) Matched–Shuffled P@1 gain, both in percentage points. Lines are seed means and bands span the seedwise minimum and maximum, not confidence intervals. The dashed five-point line is a diagnostic validity-loss reference.

GraphDiT-8M
	Matched–Shuffled		Matched–Reference			Validity
Support	$\Delta\mathrm{P}@1$	$-\Delta\mathrm{N}@10$	$\Delta\mathrm{P}@1$	$\Delta\mathrm{P}@10$	$-\Delta\mathrm{N}@10$	$L_{\mathrm{val}}$
USPTO-MR	$0.14$ [-1pt] $\pm 0.18$	$1.35$ [-1pt] $\pm 0.92$	$0.40$ [-1pt] $\pm 0.40$	$1.16$ [-1pt] $\pm 0.93$	$1.74$ [-1pt] $\pm 1.44$	$2.08$ [-1pt] $\pm 1.59$
Top15	$4.54$ [-1pt] $\pm 0.61$	$5.46$ [-1pt] $\pm 1.78$	$4.30$ [-1pt] $\pm 0.58$	$0.69$ [-1pt] $\pm 0.36$	$6.41$ [-1pt] $\pm 1.84$	$2.14$ [-1pt] $\pm 0.81$
Top30	$13.16$ [-1pt] $\pm 1.02$	$18.09$ [-1pt] $\pm 2.01$	$10.70$ [-1pt] $\pm 1.29$	$0.26$ [-1pt] $\pm 1.16$	$25.81$ [-1pt] $\pm 2.42$	$12.85$ [-1pt] $\pm 2.53$
Top45	$8.96$ [-1pt] $\pm 4.64$	$12.55$ [-1pt] $\pm 5.70$	$7.30$ [-1pt] $\pm 3.24$	$2.96$ [-1pt] $\pm 0.95$	$14.82$ [-1pt] $\pm 8.93$	$2.46$ [-1pt] $\pm 1.99$

Appendix

Table 13: All-support Stage-B replication of Table 1 , using seeds 42–44. Cells show the mean above

\pm

sample SD (

n=3

, denominator

n-1

), in percentage points. All alignment contrasts are higher-is-better; validity loss

L_{\mathrm{val}}=V_{R}-V_{M}

is lower-is-better. F marks 65M Top15, where all three runs use fallback selections that failed the original validation hard gate. Reference is fixed across Stage-B seeds.

Capacity	$(e,\alpha_{\mathrm{eval}})$	$\Delta$ P@1	$\Delta$ P@10	$\Delta$ N@1	$\Delta$ N@10	$\Delta$ Validity
8M	$(12,.50)$	$+1.25$	$+1.48$	$-2.19$	$-5.39$	$+1.12$
65M	$(14,.50)$	$+1.17$	$+0.86$	$-0.63$	$-3.28$	$-1.16$

Appendix

Table 14: Matched-validation summary for the Stage-A initialization ablation.

\Delta

is RIGS minus Direct Stage-B in percentage points. Larger

\Delta

k

, smaller

\Delta

k

, and positive

\Delta

Validity favor RIGS. The 65M row is a matched-scale validation comparison, not a test result.

Metric	RIGS M	RIGS S	Direct M	Direct S	RIGS M $-$ S	Direct M $-$ S	Direct $-$ RIGS DiD [95% CI]
P@1	24.80	12.03	23.26	11.31	$+12.77$	$+11.96$	$-0.82$ [ $-3.47,+1.89$ ]
P@3	37.09	21.56	34.70	21.27	$+15.53$	$+13.43$	$-2.10$ [ $-4.90,+0.70$ ]
P@5	39.83	25.11	37.74	25.36	$+14.73$	$+12.38$	$-2.35$ [ $-5.01,+0.37$ ]
P@10	42.60	29.37	41.39	28.73	$+13.24$	$+12.66$	$-0.58$ [ $-3.25,+2.14$ ]
N@1	2.93	13.62	5.08	12.74	$-10.69$	$-7.67$	$+3.02$ [ $+0.89,+5.19$ ]
N@3	5.46	22.42	9.76	22.49	$-16.96$	$-12.73$	$+4.23$ [ $+1.91,+6.62$ ]

Appendix

Table 15: Frozen-test Stage-A initialization difference-in-differences for USPTO–Guide(RDKit5)–8M–Top30. RIGS and Direct Stage-B use the same epoch-12 checkpoint coordinate,

\alpha_{\mathrm{eval}}=.50

, test cohort, and sampling schedule; M/S denote Matched/Shuffled. Absolute arms are case-balanced percentages; effects and DiD are percentage points. Intervals use 10,000 paired case-cluster bootstrap replicates. For P@

k

and validity, a negative Direct-minus-RIGS DiD favors RIGS; for N@

k

, a positive DiD favors RIGS because lower recovery is better.

Figure 7: Stage-B loss ablation. Full-minus-ablation effects (pp) for USPTO–Guide(RDKit5)–65M–Top30 at epoch 14 and $\alpha_{\mathrm{eval}}=.50$ . Red favors the full loss; blue favors deletion. Asterisks mark $p_{\mathrm{H}}<.05$ , Holm-adjusted over the five deletions separately for each endpoint. Complete 95% paired-bootstrap intervals and adjusted $p$ -values appear in the supplementary statistical table. This is a single-seed, validation-only comparison.

(a) Literal-prefix proposal support
Proposal	$B$	Hit@ $B$	Unique valid	Duplicate rate
Reference	5	24.38	3.40	1.55
Reference	25	50.98	15.68	6.08
Reference	100	64.80	53.53	15.23
Matched	5	18.44	2.64	0.62
Matched	25	45.35	12.59	2.42

Appendix

Table 16: Budgeted Top30 proposal and selector results. Panel (a) reports proposal support at representative budgets. Panel (b) reports all selectors used for the primary Reference and Matched pools at

B=100

. Hit, P@1, N@1, P@1/Hit, and duplicate rate are percentages; Unique valid is a mean count, and RDKit regret uses raw descriptor units.

Proposal	Head seed	Selected epoch	P@1	N@1	RDKit regret
Reference	42 (primary)	3	22.23	3.32	0.510
Reference	43	1	23.05	9.22	0.895
Reference	44	2	22.38	5.94	0.677
Matched (e12)	42 (primary)	3	19.06	0.62	0.588
Matched (e12)	43	1	21.37	1.13	0.814
Matched (e12)	44	2	19.88	0.86	0.689

Appendix

Table 17: Scalar-head seed sensitivity at

B=100

on the frozen test pools. Seed 42 is the registered primary head; seeds 43 and 44 are sensitivity runs. We select checkpoints independently on Reference validation and reuse them without adaptation across proposal sources. P@1 and N@1 are percentages.

USPTO–Guide(Struct)–8M–Top30 e22, $\alpha=.50$
Arm	P@1	P@3	P@5	P@10
Reference	30.554	49.674	57.079	64.114
Matched	35.382	53.525	58.960	63.181
Shuffled	32.065	49.348	54.902	59.775
USPTO–Guide(Struct)–65M–Top30 e26, $\alpha=1$
Reference	31.117	51.466	59.064	66.558

Appendix

Table 18: Native-cohort Structure absolute-arm results (%) on 2,229 cases and 6,752 groups. Values average groups; paired case-balanced effects appear in Table 19 .

Metric	8M, e22, $\alpha=.50$	65M, e26, $\alpha=1$
$\Delta$ P@1	+3.554 [2.491, 4.598]	+8.089 [6.910, 9.243]
$\Delta$ P@10	+3.769 [2.939, 4.573]	+5.003 [4.155, 5.853]
$-\Delta$ N@1	+3.908 [3.082, 4.755]	+7.799 [6.878, 8.753]
$-\Delta$ N@10	+3.434 [2.525, 4.341]	+18.444 [17.245, 19.651]
$\Delta$ Validity	+0.051 [-0.086, +0.190]	-0.035 [-0.278, +0.212]

Appendix

Table 19: Case-specific structural control: case-balanced Matched–Shuffled effects in percentage points with 95% case-cluster intervals. Positive

\Delta

P and

-\Delta

N indicate promotion and suppression, respectively.

Figure 8: Chemistry-audited qualitative witness for “Choose precursor sets with fewer halogen atoms.” Each column shows five manually reviewed examples selected from frequency-ranked valid unique outputs, not the unfiltered model Top-5. Evidence badges denote exact-product experiments (Exact), close-substrate experiments (Close), and reaction-class-only precedent (Class). The bottom row pairs outputs at fixed RC schedule and seed. Positive/negative labels denote frozen-set membership, independently of the evidence badges. Across all 100 draws, Reference has 21 negative and 19 positive hits at 91% validity; Guided has 0 and 24 at 85% validity.

Figure 9: Selected proofs for two qualitative planning witnesses. (A) Ambiguity-low redirects the route at unchanged depth; chemical feasibility remains unverified. (B) Halogen-low redirects the selected plan; both searches find a complete plan, and both selected plans pass the structural audit. Cards are molecule OR nodes, diamonds are reaction AND nodes, and

Xn

marks

n

halogen atoms.

Capacity	Support	$\Delta M_{+}$	$-\Delta M_{-}$	$\Delta(M_{+}-M_{-})$	$\Delta$ Validity
8M	USPTO-MR	$+0.30$ $[+0.21,+0.39]$	$+0.38$ $[+0.30,+0.46]$	$+0.67$ $[+0.53,+0.82]$	$-0.10$ $[-0.33,+0.13]$
8M	Top15	$+0.94$ $[+0.82,+1.05]$	$+1.04$ $[+0.91,+1.17]$	$+1.98$ $[+1.78,+2.17]$	$-0.06$ $[-0.21,+0.09]$
8M	Top30	$+2.27$ $[+2.10,+2.44]$	$+2.26$ $[+2.09,+2.43]$	$+4.53$ $[+4.24,+4.82]$	$-0.11$ $[-0.43,+0.22]$
8M	Top45	$+2.67$ $[+2.48,+2.85]$	$+2.69$ $[+2.50,+2.89]$	$+5.36$ $[+5.04,+5.68]$	$+0.15$ $[-0.18,+0.48]$
8M	Top60	$+1.26$ $[+1.15,+1.37]$	$+1.20$ $[+1.09,+1.32]$	$+2.46$ $[+2.28,+2.64]$	$0.00$ $[-0.20,+0.20]$
65M	USPTO-MR	$+0.12$ $[+0.04,+0.19]$	$+0.16$ $[+0.10,+0.22]$	$+0.27$ $[+0.17,+0.38]$	$+0.06$ $[-0.13,+0.24]$

Appendix

Table 20: Direct empirical sampled-mass reallocation at

B=100

. Each cell is the equal-axis, case-balanced Matched–Shuffled effect in percentage points, followed by its 95% paired-bootstrap interval. We reverse the sign of disfavored mass, so positive

-\Delta M_{-}

is favorable. The ten rows are the selected Common Top100 checkpoints registered in Table 5 .

Arm	Hit ${}_{+}@100$	P@1	P@3	P@5	P@10	N@1	N@3	N@5	N@10	Validity
		Preferred recovery $\uparrow$				Disfavored recovery $\downarrow$
Reference	35.59	3.48	7.89	11.52	18.52	3.83	7.77	10.51	18.63	73.82
Matched	35.63	4.45	10.47	14.80	21.29	0.66	1.99	3.20	5.04	69.39
Shuffled	23.91	2.93	5.90	7.97	12.62	2.42	6.13	8.83	12.23	69.26

Appendix

Table 21: Pistachio 65M Top30 five-axis results on recorded reaction alternatives. Stage-B checkpoint epoch

e=10

; residual scale

\alpha=0.50

. All entries are percentages. Hit

{}_{+}@100

measures preferred-set availability; we compute validity over raw draws. P@

k

/N@

k

use frequency-ranked unique valid outputs.

GraphDiT-8M
		Matched–Shuffled		Matched–Reference
Support	$(e,\alpha)$	$\Delta$ P@1	$-\Delta$ N@10	$\Delta$ P@1	$\Delta$ Validity
USPTO-MR	$(14,.50)$	$+0.76$ $[-0.05,+1.63]$	$+2.62$ $[+1.48,+3.77]$	$+1.35$ $[+0.56,+2.19]$	$-1.46$ $[-1.72,-1.20]$
Top15	$(12,.25)$	$+3.46$ $[+2.14,+4.84]$	$+7.03$ $[+5.50,+8.61]$	$+3.06$ $[+1.55,+4.58]$	$-2.88$ $[-3.10,-2.65]$
Top30	$(6,.25)$	$+5.37$ $[+3.69,+7.08]$	$+7.41$ $[+5.91,+8.96]$	$+4.07$ $[+2.34,+5.78]$	$-2.76$ $[-2.97,-2.55]$
Top45	$(14,.50)$	$+12.17$ $[+10.08,+14.21]$	$+19.96$ $[+17.64,+22.30]$	$+9.04$ $[+6.98,+11.05]$	$-3.62$ $[-3.90,-3.34]$

Appendix

Table 22: Instruction control under Metric4 supervision. Entries are percentage-point effects with 95% confidence intervals. Matched–Shuffled tests instruction identity; Matched–Reference measures gains and validity costs over unguided generation. Positive

\Delta

P and

-\Delta

N are favorable; negative

\Delta

Validity indicates a loss.

(e,\alpha)

gives the validation-selected Stage-B epoch and guidance scale.

Unit	Definition
Product/case	One split-local canonical target product and its candidate pool.
Graph route	One candidate precursor set that survives canonicalization, mapping, graph-capacity, and route-binding checks.
Semantic group	One case–instruction–polarity comparison over several routes.
Paired family	The two opposite-polarity semantic groups for one case and axis.
Candidate occurrence	A route appearing inside a semantic group; not a distinct base route.

Appendix

Table 23: Counting units used in the data tables.

Support	Products	Precursor sets	Avg./ product
USPTO-MR	32,584	77,110	2.37
Top15	31,912	470,308	14.74
Top30	31,912	922,977	28.92
Top45	31,912	1,360,737	42.64
Top60	31,912	1,788,168	56.03

Appendix

Table 24: USPTO-MR and synthetic expansions. Totals after preprocessing; full split counts in Table 25 .

Dataset	Support	Scale(s)	Train P/R	Validation P/R	Test P/R	Total P/R
USPTO	USPTO-MR	Both	19,022 / 41,488	6,098 / 14,740	7,464 / 20,882	32,584 / 77,110
USPTO	Top15	Both	18,595 / 274,018	5,975 / 88,142	7,342 / 108,148	31,912 / 470,308
USPTO	Top30	Both	18,595 / 537,777	5,975 / 172,969	7,342 / 212,231	31,912 / 922,977
USPTO	Top45	Both	18,595 / 793,099	5,975 / 254,984	7,342 / 312,654	31,912 / 1,360,737
USPTO	Top60	Both	18,595 / 1,042,205	5,975 / 335,110	7,342 / 410,853	31,912 / 1,788,168
Pistachio	Top30	65M	36,931 / 1,189,822	391 / 13,142	3,516 / 17,549	40,838 / 1,220,513

Appendix

Table 25: Base-generator data sizes. Each cell is usable products / accepted graph routes (P/R). Both means 8M and 65M use identical data rows; the two backbones still have separate checkpoints. USPTO Top60 reports the frozen model-training view, not the larger later annotation inventory. Pistachio’s accepted test graph view (17,549 routes) is separate from the recorded-candidate evaluation inventory (17,556 routes; Appendix G.1 ).

Suite	Signal	Construction and software	Two canonical examples
Structure	Product-local structural motifs, connectivity, fragment count, and retain/avoid criteria.	RDKit extracts candidate evidence; a frozen LLM listwise annotator proposes at most five relative criteria and route preferences. Deterministic filtering forms prefer/avoid views; RXNMapper ( Schwaller et al., 2021 ) and RDKit bind surviving routes to executable graphs.	Prefer precursors that already contain the imidazole ring directly attached to the propyl chain. Avoid precursors containing heteroatoms other than bromine and oxygen.
Metric4	mosyn.synthetic_complexity_delta , mosyn.procurement_burden , rdkit.halogen_atom_count , and rdkit.ring_count .	MOSYN joins deterministic route metrics; SCScore supplies molecular complexity and ASKCOS buyables data supplies procurement inputs. RDKit computes the two structural counts. Pool-normalized signed tails create the two polarities.	Choose disconnections that deliver a larger reduction in synthetic complexity. Choose precursor sets with a lighter procurement burden.
RDKit5	RDKit halogen-atom, carbonyl-group, total-ring, aromatic-ring, and heterocycle counts.	RDKit atom/ring APIs and fixed SMARTS compute each descriptor over the complete precursor set. Pool-normalized, tie-safe signed tails produce labels; deterministic rule-authored templates produce wording.	Choose precursor sets with fewer halogen atoms. Choose precursor sets with more aromatic rings.

Appendix

Table 26: Guidance types, construction methods, and representative canonical instructions. We compute metric values before writing instructions and do not ask the language model to estimate Metric4 or RDKit5 values.

RDKit5
Dataset	Support	Train G/R	Validation G/R	Test G/R
USPTO	USPTO-MR	60,404 / 41,488	20,610 / 14,740	25,598 / 20,882
USPTO	Top15	82,912 / 261,905	26,644 / 84,436	32,370 / 103,112
USPTO	Top30	130,880 / 528,659	42,116 / 170,286	51,522 / 208,613
USPTO	Top45	99,548 / 768,621	31,970 / 247,495	39,110 / 303,070
USPTO	Top60	134,226 / 1,046,897	43,100 / 336,212	52,564 / 410,175

Appendix

Table 27: Guidance-data sizes for the reported models. Numeric cells are semantic groups / graph-bound route keys or rows (G/R). Each USPTO row applies to both 8M and 65M; the Pistachio row applies to 65M only.

Case universe / view	Cases	Raw route records	Graph-resolved route keys	Paired families	Signed groups
Up-to-100 source test universe	7,342	733,017	666,086	29,320	58,640
Five-axis graph-bound execution view	7,269	725,812	659,056	29,314	58,628
Exact-100 eligible universe	7,320	732,000	665,146	29,262	58,524
Strict-Top100 formal cohort	1,201	120,100	107,801	1,280	2,560

Appendix

Table 28: Size accounting for the strict-Top100 common-support test cohort. We recompute every column on the named case universe: raw routes count the frozen candidate inventory, graph-resolved route keys count routes retained by the frozen graph cache, and paired families and signed groups count the corresponding five-axis semantic records.

Guided experiments	Route encoder	Dimension
USPTO 8M Top30: RDKit5, Metric4, and Structure	Pretrained reaction-pair encoder (Top30)	256
USPTO 8M Top15/Top45/Top60: RDKit5 and Metric4	Pretrained reaction-pair encoder (Top60)	256
USPTO-MR 8M: RDKit5 and Metric4	Native	256
USPTO 65M: all RDKit5/Metric4 supports and Top30 Structure	Native	512
Pistachio 65M Top30: RDKit5	Native	512

Appendix

Table 29: Frozen route encoders in the reported experiments. Native features use the Base model trained on the corresponding support.

Figure 10: Stage-B components and assembly. (a) A frozen language encoder and a trainable Stage-A-initialized projector construct reusable instruction-token memory. (b) Node queries attend to that memory through a trainable adapter whose output projection starts at zero; masking, residual scaling, and the node-state skip are explicit. (c) The frozen GraphDiT block propagates the guided node update through its existing node–edge coupling. (d) We reuse the same components at selected pre-block injection sites. The adapters inject guidance directly only into node states; edge states respond inside the frozen backbone.

Symbol	Definition
Support and supervision
$x,\ r_{i}$	Product graph and candidate precursor-set graph.
$g=(x,c,\mathcal{C}_{g},\mathcal{Y}_{g})$	Product, instruction, indexed pool, and supervision record.
$\mathcal{D}_{\mathrm{tr}},\mathcal{D}_{\mathrm{val}}$	Labeled groups for training and validation.
$K,\ K_{x},\ \mathcal{A}_{K}(x)$	Nominal support size, realized size, and retained ordered template-expansion prefix.
$\mathcal{P}_{c},\ \mathcal{N}_{c},\ q^{c}$	Preferred/avoided index sets; listwise target distribution.

Appendix

Table 30: Recurring notation across support construction, two-stage training, and evaluation.

Suite / scale	Training epochs Stage A / B	Stage-A selected epoch	Stage-B test $(e,\alpha_{\mathrm{eval}})$
Structure / 8M	30 / 30	28	$(22,.50)$
Structure / 65M	20 / 30	20	$(26,1)$

Appendix

Table 31: Structure Top30 training and validation-frozen checkpoints. Training epochs are Stage A/Stage B; the selected Stage-A epoch initializes Stage B, whose test setting is

(e,\alpha_{\mathrm{eval}})

. Both capacities use Matched-P@1-first selection.

Explore similar work

Jun 23, 2026cs.LG

Project Ariadne: Prompt-Conditioned Route Generation for Synthesis Planning

Retrosynthetic planning seeks to connect a target molecule to commercially available starting materials through a multistep route. Classical planners construct such routes by iteratively applying single-step reaction models within a search procedure; constrained variants often require specialized algorithms or architectural changes. Direct route generation reframes retrosynthesis as sequence generation, but existing direct-generation methods still train separate models for different planning specifications. We introduce Ariadne, a decoder-only route generator that represents the target, optional constraints, and route in one prompt-completion sequence. On the RetroCast/PaRoutes mkt-cnv-160 benchmark family, one 24-layer checkpoint follows route-depth and required-starting-material prompts: adding the corresponding prompt fields raises Solv-0 by 13.7 points for depth constraints and 31.2 points for required-leaf constraints. Ariadne also improves over DESP, a bidirectional search planner, on required-leaf Top-10 and Solv-0 in 24 GPU-minutes versus 6.8 GPU-hours. On standard reconstruction, Ariadne is comparable to DMS Explorer XL at about half the reported inference time. Across additional target-only benchmarks, Ariadne's clearest gains are on route-holdout reconstruction, whereas AiZynthFinder MCTS remains stronger on several Solv-0 comparisons. These results extend sequence generation from specialist retrosynthesis models to prompt-conditioned structural route generation. We release the codebase and training scripts to support further work, but do not introduce Tier-1--3 route checkers; those remain the main bottleneck before models of this kind can become useful to experimental chemists.

Anton Morgunov, Victor S. Batista

Yale University

Molecular Generation Synthesis Student-Generated Trajectories

Ask PDF

Mar 13, 2026cs.LG

RetroReasoner: A Reasoning LLM for Strategic Retrosynthesis Prediction

Retrosynthesis prediction aims to identify reactants that can synthesize a given product molecule. Although molecular large language models (LLMs) have recently shown promising results, most existing methods either generate reactants directly or provide only generic product-level analysis, without explicitly reasoning about bond-disconnection strategies that justify specific reactant choices. This paper proposes RetroReasoner, a retrosynthetic reasoning model that captures chemists' strategic disconnection-based thinking. RetroReasoner is trained with supervised fine-tuning and reinforcement learning. For supervised fine-tuning, SyntheticRetro generates structured disconnection rationales paired with reactant predictions. For reinforcement learning, a round-trip reward evaluates predicted reactants by passing them through a forward synthesis model and rewarding predictions that reconstruct the original product. RetroReasoner can also be applied to multi-step retrosynthetic planning by incorporating it into a parallelized Monte Carlo tree search framework, reducing search time while increasing the number and diversity of valid synthetic pathways. Experimental results show that RetroReasoner outperforms prior baselines, including not only molecular LLMs but also retrosynthesis-specific expert models, and generates a broader range of feasible reactant proposals, especially for challenging reaction instances. The code is available at https://github.com/KU-AGI/RetroReasoner.

Hanbum Ko, Chanhui Lee, Ye Rin Kim +4

Department of Artificial Intelligence, Korea University · Materials Intelligence Lab, LG AI Research · Department of Statistics, Korea University

Reaction Synthesis LLM Reasoning Strategies Language Modeling

Ask PDF

Aug 31, 2026cs.CL

An Agentic Retrobiosynthesis Framework with Learned Frontier Selection

Large language models are increasingly used as agents for multistep retrosynthesis, raising the question of how much their search policy contributes independently of the underlying reaction model. We investigate this question in a biological setting through rule-based retrobiosynthesis: a deterministic biochemical engine generates the same validated transitions for every method, searching for routes that terminate in metabolites available to an \emph{Escherichia coli} chassis, while the policy only selects which frontier molecule to expand next. Prompted and LoRA-tuned Qwen2.5-7B policies use a strict choice-only interface. The fine-tuned policy reaches

65\pm1

% solve rate at 10 expansions on LASER versus 59% for MCTS, and at 200 expansions reaches

78\pm1

% versus 75% on LASER,

88\pm3

% versus 80% on the RetroPath RL Golden benchmark, and

63\pm2

% versus 45% on the BioNavi-NP benchmark. Fine-tuning also consistently outperforms direct prompting. These results show that route-supervised frontier selection can improve budgeted search without altering biochemical generation, although performance remains dependent on frontier construction and reaction ranking.

Philippe Meyer, Guillaume Gricourt, Thomas Duigou +2

1Université Paris-Saclay, INRAE, AgroParisTech, Micalis Institute, Jouy-en-Josas, France

Large Language Model Policy Optimization Search-Augmented Reasoning Bioinformatics Frontiers

Ask PDF