Papers for

protein engineering teams

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Antibody developability prediction improves with shared structure and language models

Multi-Assay Antibody Developability Prediction with Shared Structure-Language Representations

Abstract: Early assessment of antibody developability is critical in drug discovery, but computational approaches remain limited by the scarcity of experimental data. We explore two strategies for working within that limit. First, we designed per-residue physicochemical descriptors encoding the local structural environment, including new terms such as directional charge asymmetry and screened electrostatic potential. These feed a convolutional network, combined with protein language model (PLM) embeddings in a two-stream architecture: the two are encoded and predicted independently and merged only at the prediction stage. Second, a shared trunk with per-assay heads passes information between related assays. Both are implemented in the Multi-Assay Developability Model (MADM) and evaluated on eight assays: HIC, heparin affinity (HAC), AC-SINS, and polyspecificity ELISA against BVP, dsDNA, insulin, ovalbumin and soluble membrane proteins (SMP), with labels from 29 open sources. Structural descriptors match or exceed PLM embeddings (AbLang2, ESM2) on several assays; combining the streams gives a small but consistent gain in accuracy and reduces seed-to-seed variance. Multi-assay training raises Spearman rho for ovalbumin and SMP by about 0.1, where assays correlate and share sequence space. Under cluster-split cross-validation the strongest results are for HAC, ovalbumin, insulin, SMP and HIC (rho = 0.70-0.82); AC-SINS, BVP and dsDNA are weaker (0.50-0.55). In these three, sequence clusters largely coincide with data source, so a held-out fold measures transfer to differently calibrated measurements as well as to new sequences. On the 80-antibody GDPa3 hold-out, performance falls by rho around 0.1 for all but SMP (0.22). Transfer to the VHH panel of Ginkgo GDPa5 gives comparable results, and scFv validation against binary SMP labels yields F1 = 0.89, AUC = 0.95.

Fri 13 NovMachine Learning
The gist
Predicting which antibodies will work well as medicines early on is hard because there isn’t much experimental data. The authors combined two types of information: detailed 3D structural features of antibodies and language-based models that analyze their sequences. Their combined model learns from multiple related tests simultaneously, improving accuracy for several important antibody properties. This helps scientists better guess which antibodies can become drugs before costly lab work.
Open → 2611.18846v1

Probabilistic steering improves antibody design efficiency and diversity

Probabilistic Steering of Generative Priors for Lab-in-the-Loop Antibody Optimization

Abstract: Therapeutic antibody development is often framed as comprising two discrete stages: initial discovery of novel binders followed by iterative optimization to refine these hits into therapeutic-grade molecules. Models and methods often reflect these stages, with generative structure models applied to discovery and probabilistic black-box optimization methods applied to refinement. Here we propose to unify these stages by smoothly transitioning a generative discovery model into iterative optimization, allowing the handoff between the structure-based prior and supervision from wet lab data to be guided by the posterior uncertainty of a probabilistic surrogate. We show that this yields improvements in optimization efficiency in the low-N regime, as well as improved diversity amongst the optimized molecules.

Mon 2 NovMachine Learning
The gist
Developing therapeutic antibodies often involves two separate steps: first finding new candidates and then improving them in the lab. The authors show how combining computer models that generate new antibody structures with lab data in a smooth way can make this process more efficient. Their method uses uncertainty in the model’s predictions to decide how much to trust computer-generated ideas versus real lab results. This approach helps find better and more varied antibodies especially when only a small amount of lab data is available.
Open → 2611.03844v1

AI structure predictions improve filtering for antibody design tests

Revisiting Structure-Based Filtering for AI-Designed Antibodies: Opportunities, Limits, and What Matters

Abstract: AI antibody design can generate far more candidates than can be experimentally tested, making post-generation filtering a central bottleneck. We study whether structure-prediction outputs can help prioritize AI-designed antibodies for experimental validation. We evaluate this problem on 4,184 experimentally characterized antibody-antigen pairs, combining newly generated wet-lab measurements with curated affinity data across diverse antigens and assay settings. We introduce two complementary filtering signals: Regional Interaction Contrast (RIC), which captures support for the predicted antibody-antigen interface, and Non-CDR Antibody Confidence (NAC), which measures antibody confidence outside the CDRs. Across AlphaFold 3, Protenix-v2, and ESMFold2, we find a striking difference in what information matters for filtering. Interface evidence is strongest when comparing candidates designed for different antigens, whereas antibody-level reliability is substantially more informative for ranking candidates within the same antigen. In particular, non-CDR antibody confidence consistently outperforms whole-antibody confidence across predictors, revealing an unexpected source of filtering signal beyond the predicted interface. Finally, we examine the information encoded in the predictor outputs themselves. Predicted structures contain substantial but incomplete binding-related signal, revealing both the opportunities and limits of structure-based filtering: better metrics can exploit current predictors more effectively, but their performance ultimately remains constrained by what those predictors can distinguish.

Fri 16 OctComputational Engineering, Finance, and Science
The gist
AI methods can create lots of new antibody designs, but it’s hard to decide which ones to test in the lab. The authors studied if predicted 3D structures of antibodies help pick the best candidates. They found that details about the antibody-antigen interface help when choosing among designs for different targets, while confidence in antibody regions outside key parts helps when comparing designs for the same target. Their work shows current prediction tools provide useful but limited clues for selecting promising antibodies.
Open → 2610.23136v1

Adversarial attacks can poison protein design data and outputs

Poisoning the corpora of protein design systems during pre-training and retrieval

Abstract: Protein foundation models increasingly support the generation of novel biological sequences, yet their reliance on large public datasets creates opportunities for adversarial manipulation of the material they learn from and retrieve. We investigate two data-only poisoning attacks against protein design systems, targeting the pre-training corpus of autoregressive protein language models and the retrieval corpus of a text-guided protein generator. For the first, we pre-trained ProtGPT2 and ProGen2 variants of 300 million parameters, introducing poisoned protein sequences containing one of two triggers followed by one of three payloads. Across all poisoned runs, both architectures acquired highly reliable trigger-dependent generation under uniform and ramped poisoning schedules. Under ramped exposure, the earliest majority activation occurred after only 200 contaminated examples, while for the uniform schedule earliest activation occurred around 5000 poisoned samples. For the second attack, we inserted records into a retrieval corpus system without modifying its weights or decoding procedure. A single query-aware record captured a top-16 retrieval position for 58\% of templated requests and 97\% of free-text requests. Conditional on successful poisoned retrieval, 62\% of requests produced altered sequences, while 19-21% acquired the inserted domain by profile-HMM homology. A universal query-unaware attack using 1,500 records also achieved substantial retrieval capture and domain transfer. Together, these findings demonstrate that adversarial protein sequence contributions can establish persistent, sequence-specific behaviour during pre-training or redirect protein generation at inference, while leaving ordinary model behaviour largely preserved. Our work establishes the integrity and biological provenance of protein sequence data as critical safeguards for generative protein design systems.

Thu 15 OctMachine Learning
The gist
Protein design tools learn from large collections of protein sequences to create new proteins. The authors show that attackers can sneak in harmful or misleading sequences during the training or retrieval phases, which makes the models produce specific altered proteins when triggered. These attacks can work with surprisingly few poisoned examples and often go unnoticed because the models still behave normally otherwise. This highlights the need to carefully check and secure protein data used in these systems.
Open → 2610.21200v1

Protein model ESM-2 retains unclear levels of genetic interaction information

ORBIT-FMIB: Tracking Order-Resolved Epistatic Information Through ESM-2

Abstract: Protein foundation models support mutation-effect and structural prediction, but predictive performance alone does not reveal which forms of biological interaction information remain accessible through model depth. We ask whether ESM-2 retains higher-order epistatic information as strongly as first- and second-order information across its representation hierarchy, introducing ORBIT-FMIB, a diagnostic framework combining Walsh-based interaction decomposition with subset-conditioned neural dependence estimation. The method is validated on synthetic landscapes with known interaction structure before being applied to the dense four-site GB1 fitness landscape using frozen ESM-2 representations. An initial production run suggested ESM-2 retains higher-order epistatic information less well than lower-order information ($Δ_{\mathrm{HO-LO}}=-0.107$). An independent replication of the complete measurement grid, under matched GPU hardware and identical critic seeds, substantially reduced this contrast ($Δ_{\mathrm{HO-LO}}=-0.017$), and its sign was unstable across otherwise-defensible evaluation-pairing choices applied to the same trained critics ($-0.011$ to $+0.015$). We therefore do not currently have robust evidence that ESM-2 selectively loses higher-order epistatic information, nor that retention is equal across orders; the directional question remains open. The measurement protocol itself, including its documented removal of a positional-subset shortcut in pooled critics, remains validated and is unaffected by this finding. ORBIT-FMIB is offered as a diagnostic framework for probing interaction structure in protein foundation models; this study's own replication result illustrates why such probing requires adequately-powered reproducibility checks before its output is treated as a biological finding.

Wed 30 SeptMachine Learning
The gist
Predicting how protein mutations interact is important for biology but tricky to measure from AI models. The authors created a new test, ORBIT-FMIB, to check if a protein model called ESM-2 keeps information about complex mutation effects. Their tests showed inconsistent results about whether ESM-2 remembers higher-level mutation interactions well. This means it is still unknown if the model is better at learning simple or complex mutation effects. The test itself is reliable and can help others study protein models in the future.
Open → 2610.00672v1

PhiFold generates protein shapes and how they move together

PhiFold: Towards Dynamic Protein Design with Physics-Structured Covariance Modeling

Abstract: Protein design is moving beyond structural correctness toward function-aware design, yet existing generative models typically treat dynamics as a downstream property estimated through simulation or prediction after structure generation. Using MD trajectories as a generative target is also undesirable because stochastic, path-dependent trajectories over-specify the underlying equilibrium ensemble. We introduce PhiFold, a framework for jointly generating protein backbones and their second-order dynamics, represented by residue-displacement covariance. Rather than predicting the quadratically sized full covariance, PhiFold decomposes dynamics into three interpretable components: local flexibility, a low-rank collective-motion representation, and residue-wise collective participation. These components are assembled into a positive-definite covariance matrix with exact marginal consistency, yielding a compact and physically constrained representation of equilibrium dynamics. Across generated proteins, PhiFold improves recovery of local fluctuations and long-range residue coupling while remaining competitive on dominant collective-motion subspaces. It further enables bidirectional control of residue flexibility while preserving backbone designability. By unifying structure generation with an explicit representation of equilibrium dynamics, PhiFold lays a foundation for designing proteins not only by how they look, but also by how they move.

Sat 26 SeptArtificial Intelligence
The gist
Designing proteins with the right shape is important, but knowing how they move is even more crucial for their function. The authors introduce PhiFold, a method that creates protein backbones while also modeling their natural movements through a special kind of matrix called a covariance matrix. Instead of predicting complex data directly, PhiFold breaks it down into simpler, understandable parts to represent local flexibility and collective motions. This method helps design proteins not only by their structures but also by their dynamics, potentially improving how proteins work.
Open → 2609.32309v1