iGC: Antibody Optimization via Reinforcement Learning over a Learned Somatic Hypermutation Prior

Thao Nguyen; Robert G. Alberstein; Ewa Nowara; Homa MohammadiPeyhani; Holly R. Steach; Sarah A. Robinson; Yuval Shechter; Donald W. Lee

cs.LGNov 3, 2026

iGC: Antibody Optimization via Reinforcement Learning over a Learned Somatic Hypermutation Prior

Authors: Thao Nguyen, Robert G. Alberstein, Ewa Nowara, Homa MohammadiPeyhani, Holly R. Steach, Sarah A. Robinson, Yuval Shechter, Donald W. Lee,

Organizations: University of Illinois at Urbana-Champaign · Genentech · F. Hoffmann-La Roche Ltd.

Abstract

Antibody affinity maturation is a natural optimization process in which B cells acquire mutations through somatic hypermutation (SHM), undergo selection for improved antigen binding, and iteratively evolve toward higher affinity. Computationally modeling this process requires capturing both biologically plausible mutations and antigen-specific selection. However, existing mutation models do not directly optimize binding, while affinity optimization often neglects realistic mutation trajectories. We introduce iGC (in silico Germinal Centers), a two-stage reinforcement learning (RL) framework for antigen-specific antibody optimization. Stage 1 learns an SHM mutation prior from parent-child pairs in B-cell lineage trees and adapts it to a target affinity-maturation dataset. Stage 2 initializes an RL policy from this prior and optimizes antigen binding with Kullback-Leibler (KL) regularization toward the frozen prior. To prevent reward exploitation, including repetitive high-reward substitutions and structurally tolerated framework mutations, we introduce a Reward-Tax Meta-Gradient objective. RewardNet smooths the affinity landscape, while TaxNet penalizes reward overexploitation, jointly defining a learned meta-reward for policy optimization. We evaluate iGC on mutation prediction, SHM consistency, affinity optimization, and mutation diversity. The mutation prior more than doubles the strongest baseline in mutation-site prediction. Following RL optimization, iGC maintains SHM sequence structure. It also achieves a strong exploration-exploitation balance, yielding competitive reward-diversity products across candidate-selection cutoffs. These results demonstrate that iGC integrates biologically plausible mutation generation with antigen-specific affinity selection, providing an in silico framework for iterative antibody affinity maturation.

Figures & tables

Figure 1: (a) Affinity maturation in germinal centers as a repeated mutation–selection process. (b) Structure of an antibody and the heavy-chain variable region (VH).

Figure 2: Overview of the two-stage iGC framework. Stage 1 learns and domain-adapts an SHM mutation prior, and Stage 2 optimizes the resulting policy for antigen binding while remaining regularized toward that prior. At inference, the learned policy optimizes an input antibody by rolling out successive mutations under a fixed mutation budget.

Figure 3: Reward landscape across antibody structural regions. (a) Raw GCReplay rewards are imbalanced across regions, driving hotspot and safe-region policy collapse. (b) The learned RTMG meta-reward smooths extreme peaks and produces a more balanced reward landscape.

Model	pos@1 ( $\uparrow$ )	pos@3 ( $\uparrow$ )	pos@5 ( $\uparrow$ )	aa@1 ( $\uparrow$ )	aa@3 ( $\uparrow$ )	aa@5 ( $\uparrow$ )	Joint ( $\uparrow$ )
S5F	2.433 $\pm$ 0.172	7.813 $\pm$ 0.293	12.237 $\pm$ 0.457	26.130 $\pm$ 0.569	64.310 $\pm$ 0.601	84.953 $\pm$ 0.331	0.637 $\pm$ 0.057
Thrifty	4.257 $\pm$ 0.183	10.633 $\pm$ 0.249	15.533 $\pm$ 0.230	32.777 $\pm$ 0.367	68.747 $\pm$ 0.450	86.813 $\pm$ 0.052	1.395 $\pm$ 0.066
DASM selection-only	1.070 $\pm$ 0.149	3.960 $\pm$ 0.163	6.947 $\pm$ 0.169	18.043 $\pm$ 0.548	38.497 $\pm$ 0.649	52.203 $\pm$ 0.790	0.194 $\pm$ 0.032
DASM (mutation $\times$ selection)	5.307 $\pm$ 0.058	13.133 $\pm$ 0.393	19.197 $\pm$ 0.382	40.561 $\pm$ 0.555	75.737 $\pm$ 0.365	89.435 $\pm$ 0.225	2.153 $\pm$ 0.052
DeepSHM	5.020 $\pm$ 0.302	11.747 $\pm$ 0.253	17.467 $\pm$ 0.299	38.850 $\pm$ 0.653	74.250 $\pm$ 0.644	88.470 $\pm$ 0.122	1.949 $\pm$ 0.105
iGC (Stage 1 pretraining)	11.929 $\pm$ 0.141	25.169 $\pm$ 0.217	33.836 $\pm$ 0.238	40.777 $\pm$ 0.323	75.552 $\pm$ 0.321	90.027 $\pm$ 0.188	4.864 $\pm$ 0.052

Table 1: Prediction accuracy of mutation position and amino acid identity for different SHM models. Joint accuracy is the fraction of examples where both the predicted mutation position and amino acid substitution are correct. Values are reported as mean

\pm

standard error over three data splits. Arrows indicate the preferred direction for each metric.

Figure 4: Stage 1 SHM mutation prediction on the Tang test set: top-1 mutation-site accuracy (left), top-1 amino-acid accuracy (middle), and joint accuracy (right) for each model. Bars show the mean over three data splits (in %), with error bars indicating the standard error; values are the same as in Table 1 .

Model	DMS	#Dist	Prod	DMS	#Dist	Prod	DMS	#Dist	Prod	DMS	#Dist	Prod
	Top 100			Top 30			Top 20			Top 10
SHMReplay	$0.514$	257.6	132.4	$0.878$	99.3	87.2	$0.990$	70.7	70.0	$1.164$	38.5	44.8
	(0.018)	(2.3)	(5.3)	(0.033)	(1.5)	(3.4)	(0.040)	(0.7)	(3.1)	(0.047)	(0.4)	(1.7)
CoSiNE	$0.019$	195.3	3.7	$0.990$	70.4	69.7	$1.155$	50.0	57.8	$1.372$	27.5	37.7
	(0.078)	(17.0)	(15.7)	(0.073)	(2.5)	(5.6)	(0.077)	(2.3)	(4.5)	(0.058)	(1.6)	(2.7)
Adapted SHM Prior	$0.185$	166.9	30.9	$1.045$	62.8	65.6	$1.195$	44.4	53.1	$1.414$	25.0	35.4

Table 2: Stage 2 benchmarking results averaged equally over mutation budgets

k=2

7

, with the same number of generated sequences at each budget, and reported over three data splits. Each model’s mean row is followed by a parenthesized standard-deviation row across splits. DMS is mean reward, #Dist is the number of distinct mutation identities, and

\mathrm{Prod}=\mathrm{DMS}\times\#\mathrm{Dist}

. Higher is better for all metrics. Bold and underlined Prod values denote the best and second-best means, respectively. ∗ denotes RL-based methods initialized from the Adapted SHM Prior.

Appendix figures & tables11 assets

Supplementary material from the paper’s appendix.

Appendix

	WRC	WA	GYW
Real mutation positions	26.9%	24.9%	18.7%
iGC predicted positions	57.0%	1.0%	11.1%
Random background	2.0%	19.4%	4.1%

Appendix

Table 3: Canonical SHM motif frequencies for the final iGC policy after Stage 2 fine-tuning.

$k$	Unseen positives rediscovered	Total unseen positives	Rediscovery rate
2	54	169	32.0%
3	41	375	10.9%
4	15	255	5.9%

Appendix

Table 4: Rediscovery of unseen positive-reward sequences in the held-out test set.

RewardNet input	Avg. DMS reward ( $\uparrow$ )	Avg. #distinct mut. ( $\uparrow$ )
cum_dms + pos_onehot	0.526	129.3
cum_dms + mut_count	0.743	116.6
cum_dms + mut_count + pos_onehot + region_onehot	1.134	108.0

Appendix

Table 5: RewardNet input selection. The chosen configuration is shown in bold.

Floor	Avg. DMS reward ( $\uparrow$ )	Avg. #distinct mut. ( $\uparrow$ )
$-6.0$	0.528	155.9
$-5.5$	0.263	178.1
$-5.0$	0.411	129.4
$-4.5$	1.142	106.3
$\mathbf{-4.0}$	1.134	108.0
$-3.5$	0.880	127.1

Appendix

Table 6: TaxNet log sequence-score floor selection over mutation budgets

k=1

7

. The selected floor is shown in bold.

Pipeline	pos@1	aa@1	joint	FR4%	CDR2%
P1: No Stage 1	0.296	0.417	0.124	7.4%	33.7%
P2: Stage 1 pretrain only	0.286	0.512	0.146	26.6% ‡	14.8%
P3: Full Stage 1	0.301	0.539	0.162	1.2%	21.6%

Appendix

Table 7: Pipeline ablation on GCReplay test set. FR4% and CDR2% are argmax region distributions; empirical rates are 1.4% and 26.7% respectively.

Figure 5: Mean attention weight per transformer layer as a function of relative sequence position to the predicted mutation site (position 0, dashed red), aggregated across the Tang BCR test set. Yellow bands highlight the $\pm$ 5 position window. All layers consistently attend most strongly to the local neighborhood of the mutation site, consistent with AID’s known short-motif targeting mechanism.

	WRC	WA	GYW
Real mutation positions	26.9%	24.9%	18.7%
Adapted SHM Prior predicted positions	54.4%	1.8%	42.1%
Random background	2.0%	19.4%	4.1%

Appendix

Table 8: Canonical SHM motif frequencies for the Stage 1 Adapted SHM Prior.

Training phase	From-scratch	Scaled	AbLang2
	(4.8M)	(14.3M)	(44.4M)
Stage 1: Tang pretraining	4.864%	4.86%	4.97%
Stage 1: GCReplay adaptation	15.62%	14.19%	15.87%
Stage 2: GCReplay RL	15.20%	9.47%	13.37%

Appendix

Table 9: Backbone ablation across training phases of the two-stage pipeline. Entries report joint@1 accuracy (%).

Figure 6: Stage 1 scaling law: test-set accuracy versus number of training trajectories for position (left), amino-acid (middle), and joint (right) prediction. Each point is a separate model trained on one PCP dataset (Ford, Rodriguez, Jaffe, Tang; the Jaffe point is labeled “Wyatt” in the right panel). Dashed lines are power-law fits $\text{acc}=a\,n^{b}$ extrapolated beyond the largest dataset; fitted exponents are given in the legends.

Figure 7: Reward–tax meta-gradient (RTMG) architecture. RewardNet converts the measured DMS reward and mutation context into a learned benefit, while TaxNet penalizes recurrent or SHM-accessible sequence outcomes according to their evolutionary sequence-probability score. Their difference defines the meta-reward used for a differentiable update of the probe policy; feedback from the probe policy then trains the reward and tax networks to improve affinity while preserving mutation diversity. The deployed policy is updated using the resulting meta-reward.

Component	Hyperparameter	Value
RTMG objective	Entropy weight $\lambda_{H}$	$3.0$
	Reward fidelity weight $\lambda_{R}$	$1.0$
	Tax fidelity weight $\lambda_{T}$	$1.0$
	Net-reward fidelity weight $\lambda_{\mathrm{net}}$	$1.0$
	RewardNet cap $C_{R}$	$5.0$
	RewardNet pinned mean $\mu_{\mathrm{reward}}$	$0.9937$

Appendix

Table 10: Hyperparameters for the full RTMG model and mutation-policy Transformer.

Protein Design

Protein design uses computational and experimental methods to create or modify protein sequences and structures for desired functions.

See all 107 papers

Conditionally Site-Independent Neural Evolution of Antibody Sequences
Feb 21, 2026Stephen Zhewen Lu, Aakarsh Vermani, Kohei Sanno +4Markov Chains Molecular Generation
University of California, Berkeley · Mila - Québec AI Institute · Université de Montréal +4
AbICL: In-Context Learning for Antigen-Specific Antibody Affinity Ranking
Jul 7, 2026Zhiyuan Chen, Jing Hu, Junzhe Wang +4Learning to Rank Meta-Learning
Henlius Shanghai, China
AbLWR:A Context-Aware Listwise Ranking Framework for Antibody-Antigen Binding Affinity Prediction via Positive-Unlabeled Learning
Apr 13, 2026Fan Xu, Zhi-an Huang, Haohuai He +5Learning to Rank
Department of Data Science and Artificial Intelligence, The Hong Kong Polytechnic University, Hong Kong SAR, China · Department of Computer Science, City University of Hong Kong (Dongguan), Dongguan, China · School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, Guangdong, China +1

Feb 21, 2026J/K move · Enter open · S save

Conditionally Site-Independent Neural Evolution of Antibody Sequences

Stephen Zhewen Lu, Aakarsh Vermani, Kohei Sanno, Jiarui Lu+3

University of California, Berkeley · Mila - Québec AI Institute · Université de Montréal+4

Markov Chains Molecular Generation Protein Fitness Prediction

Open paper AskPDF arXiv

Common deep learning approaches for antibody engineering focus on modeling the marginal distribution of sequences. By treating sequences as independent samples, however, these methods overlook affinity maturation as a rich and largely untapped source of information about the evolutionary process by which antibodies explore the underlying fitness landscape. In contrast, classical phylogenetic models explicitly represent evolutionary dynamics but lack the expressivity to capture complex epistatic interactions. We bridge this gap with CoSiNE, a continuous-time Markov chain parameterized by a deep neural network. Mathematically, we prove that CoSiNE provides a first-order approximation to the intractable sequential point mutation process, capturing epistatic effects with an error bound that is quadratic in branch length. Empirically, CoSiNE outperforms state-of-the-art language models in zero-shot variant effect prediction by explicitly disentangling selection from context-dependent somatic hypermutation. Finally, we introduce Guided Gillespie, a classifier-guided sampling scheme that steers CoSiNE at inference time, enabling efficient optimization of antibody binding affinity toward specific antigens.

Abstract

Protein Design

Explore similar work

Conditionally Site-Independent Neural Evolution of Antibody Sequences

AbICL: In-Context Learning for Antigen-Specific Antibody Affinity Ranking

AbLWR:A Context-Aware Listwise Ranking Framework for Antibody-Antigen Binding Affinity Prediction via Positive-Unlabeled Learning