Definitions of every technical term used across the WeaveSeq platform — from computational metrics to experimental validation. Each entry is cross-referenced to related terms for fast navigation.
pLDDT is AlphaFold's primary self-assessment metric, computed as a local distance difference test on the Cα trace. Residues scoring above 0.9 are typically modelled at high accuracy (comparable to X-ray crystallography), while values below 0.5 may indicate intrinsically disordered regions. In binder design, per-residue pLDDT reveals which interface positions are structurally reliable and which carry high uncertainty; this information directly informs candidate ranking and lead selection.
See also: i_pTM, iP AE, AlphaFold Confidencei_pTM (interface predicted TM-score) measures how well AlphaFold-Multimer predicts the relative orientation of two interacting chains. A score above 0.8 suggests the interface is modelled with high confidence, while values below 0.5 indicate substantial uncertainty in the docking geometry. Unlike global pTM which assesses the entire complex, i_pTM isolates the interface, the region that determines binding specificity and affinity. It is one of the primary metrics WeaveSeq uses to gate candidates between computational design and experimental validation.
See also: pLDDT, iP AE, Interface Contacts, Multi-State Validation (MSV)iP AE (interface predicted aligned error) is the interface-specific subset of the full PAE matrix. It reports the expected positional error for each inter-chain residue pair, measured in Ångströms. Low iP AE values (< 5 Å) indicate a well-resolved interface; values above 15 Å suggest the relative positioning is unreliable. WeaveSeq reports iP AE alongside pLDDT and i_pTM to give a complete picture of structural confidence: a design may have high pLDDT within each chain but poor interfacial accuracy. All three are required to pass the advancement gate.
See also: pLDDT, i_pTM, Interface Contacts, AlphaFold ConfidenceAlphaFold produces several confidence metrics alongside the predicted 3-D coordinates: pLDDT (per-residue), PAE (pairwise residue error), pTM (global template modelling), and i_pTM (interface-specific). These are NOT experimental validation; they are model-internal estimates of prediction quality. However, extensive benchmarking against experimentally determined structures (CASP, CAMEO) has shown strong correlation, making them reliable enough to serve as primary triage metrics in computational design pipelines. WeaveSeq's AffiniBind workflow uses a composite threshold: binder pLDDT ≥ 0.85, i_pTM ≥ 0.75, and iP AE ≤ 5 Å for advancement.
See also: pLDDT, i_pTM, iP AE, Multi-State Validation (MSV)The PAE matrix is an N×N heatmap where position (i, j) shows AlphaFold's expected error for the distance between residue i and residue j. Low PAE values (dark green in the standard colouring) indicate high-confidence relative positioning; high values (white/red) indicate uncertainty. While pLDDT assesses per-residue accuracy, PAE captures the model's confidence in the relative domain orientation: a structure can have high per-residue pLDDT but poor PAE if domain packing is uncertain. The interface-specific subset of PAE, termed iP AE, is a key advancement metric in AffiniBind.
See also: iP AE, pLDDT, AlphaFold Confidence, pTM (Predicted Template Modeling Score)pTM (predicted TM-score) is AlphaFold's whole-complex accuracy estimate, computed from the predicted pairwise errors. A pTM above 0.9 indicates high-confidence overall topology; values below 0.5 suggest the global fold is unreliable. For protein–protein complexes, the interface-specific variant i_pTM is more informative than global pTM: a complex can have excellent pTM if individual chains are well-folded even when the interface is poorly resolved. AffiniBind uses i_pTM rather than global pTM as its primary advancement metric for that reason.
See also: i_pTM, pLDDT, AlphaFold Confidence, PAE (Predicted Aligned Error)dSASA measures how much surface area is buried when a binder docks to its target. Larger buried surface areas generally correlate with higher binding affinity, though the relationship is not strictly linear. In WeaveSeq's pipeline, dSASA is one component of the interface metrics panel; designs with dSASA below 600 Ų are flagged as having insufficient interfacial contact, regardless of their pLDDT or i_pTM scores. Typical high-affinity protein–protein interfaces bury 1,200–2,000 Ų.
See also: Interface Contacts, Shape Complementarity (S_c), Binding Affinity (K_D)Interface contact count is a direct structural measure of how many residue–residue interactions stabilize a predicted binder–target complex. Higher contact counts generally indicate more extensive interfaces, but too many contacts concentrated in a single patch can indicate non-specific binding. AffiniBind reports interface contacts alongside contact probability (the fraction of AlphaFold-Multimer model seeds that predict each contact) to distinguish robust, consistently predicted contacts from spurious ones that appear in only a subset of models.
See also: dSASA (Delta Solvent-Accessible Surface Area), Shape Complementarity (S_c), Binding Affinity (K_D), Interface ResidueShape complementarity quantifies the geometric fit between a binder and its target. Values above 0.65 are typical for biologically relevant interfaces. Perfect shape complementarity (1.0) is neither expected nor desirable; biological interfaces require some conformational flexibility for binding and release. AffiniBind reports S_c alongside dSASA and contact counts, weighted such that moderate complementarity (0.55–0.70) with high dSASA is preferred over perfect complementarity with minimal buried surface area.
See also: dSASA (Delta Solvent-Accessible Surface Area), Interface Contacts, Binding Affinity (K_D)Binding affinity is THE fundamental performance metric in binder development. Typical therapeutic antibodies achieve K_D values in the nanomolar (10⁻⁹ M) to picomolar (10⁻¹² M) range. AffiniBind predicts structural metrics that correlate with affinity (dSASA, contact count, interface hydrophobicity) but does not directly compute K_D; experimental validation via SPR or BLI is required to confirm binding. The platform's multi-state validation module partially addresses this by identifying designs likely to maintain specificity in complex backgrounds, but wet-lab measurement remains the gold standard.
See also: dSASA (Delta Solvent-Accessible Surface Area), Multi-State Validation (MSV), SpecificityRelated reading: The Architecture of De Novo Protein BindersMulti-State Validation is AffiniBind's computational selectivity filter. Each binder design is folded by AlphaFold-Multimer in complex with the target (positive state) and with one or more off-target proteins (negative states (e.g. BSA, streptavidin)). A good binder should show strong binding metrics (high i_pTM, high interface contacts, low iP AE) against the target and weak or no binding against decoys. Designs that bind non-specifically to negative states are flagged or eliminated. The MSV module reports a composite rank score that balances affinity (positive-state metrics) with specificity (positive-to-negative margin).
See also: Specificity, Binding Affinity (K_D), i_pTM, Interface ContactsSpecificity is what separates a useful reagent from a noisy one. A high-affinity binder that also binds BSA, streptavidin, and half the proteome is worthless for most applications. AffiniBind evaluates specificity computationally through Multi-State Validation: each design's interface metrics against the target are compared to its metrics against decoy proteins. The specificity margin (the gap between positive-state and negative-state scores) is a key advancement criterion. Designs with margins below 0.1 (positive indistinguishable from negative) are rejected even if their absolute metrics are excellent.
See also: Multi-State Validation (MSV), Binding Affinity (K_D), Cross-ReactivityRelated reading: Geometric Complementarity in Interface EngineeringCross-reactivity is the practical consequence of insufficient specificity. In diagnostics, it produces false positives. In therapeutics, it causes off-target toxicity. In biosensing, it destroys signal-to-noise ratio. AffiniBind addresses cross-reactivity proactively by including explicit negative states in the computational screen, rather than discovering the problem after weeks of wet-lab work. The MSV module's negative-state metrics (BSA_i_pTM, streptavidin_i_pTM) are surfaced directly in the dashboard so engineers can inspect which decoys pose the greatest cross-reactivity risk for a given design.
See also: Specificity, Multi-State Validation (MSV)De novo binder design represents a paradigm shift from traditional antibody discovery. Instead of immunizing animals or screening phage libraries, the binder is computationally generated to complement a specific target epitope. AffiniBind's approach uses inverse folding (ProteinMPNN) conditioned on target structure, followed by AlphaFold-Multimer to predict complex structures and assess binding. This sequence-first strategy enables exploration of sequence space that is not constrained by natural immune repertoires, opening the possibility of binders to targets that have resisted traditional approaches (e.g., recessed epitopes, highly conserved surfaces, or toxic targets).
See also: Scaffold Protein, Epitope, Interface DesignIn binder design, the scaffold is the structural chassis that holds the binding residues in the correct 3-D orientation to complement the target epitope. Choosing the right scaffold is critical: it must be stable, well-expressed, monomeric (no self-association), and free of immunogenic or aggregation-prone sequence motifs. AffiniBind evaluates scaffolds through a developability panel that screens for oligomerization risk, exposed hydrophobic patches, protease motifs, and charge clustering, all before a single residue of the binding interface is designed.
See also: Developability, De Novo Binder Design, Interface DesignRelated reading: Scaffold Engineering for Multispecific BindingComputational binding energy means nothing if the protein aggregates in the expression host. AffiniBind's developability panel scores each design on: oligomerization risk (tendency to self-associate), exposed hydrophobicity (aggregation propensity), charge clustering (electrostatic repulsion), buried unsatisfied polar atoms (stability cost), packing density (fold quality), and protease motif exposure (expression half-life). These scores are normalized and combined into a composite robustness score that gates advancement alongside binding metrics.
See also: Scaffold Protein, Aggregation, Composite Robustness ScoreInterface design is the core engineering problem in binder development. The interface must satisfy multiple constraints simultaneously: geometric complementarity (shape fit), chemical complementarity (hydrophobic/hydrophilic matching), electrostatic compatibility (charge pairing without repulsion), and conformational stability (rigid, pre-organized geometry that minimises the entropic cost of binding). AffiniBind's generative engine samples thousands of sequence variants at interface positions, evaluates each against these constraints, and ranks candidates by a composite score that weights binding quality against developability risk.
See also: Shape Complementarity (S_c), dSASA (Delta Solvent-Accessible Surface Area), Epitope, Binding HotspotRelated reading: Geometric Complementarity in Interface EngineeringSequence diversity is a critical quality metric in binder library design. If the top 100 ranked candidates are all minor variants of the same scaffold with one or two mutations, testing all 100 wastes resources. AffiniBind's diversity selection stage (pipeline stage 04) clusters candidates by sequence identity and selects representatives from each cluster, ensuring the experimental pool spans distinct sequence and structural solutions. The diversity threshold is configurable (typically 70% sequence identity is used as the clustering cutoff), so the pipeline can be tuned for exploration (lower threshold) or refinement (higher threshold).
See also: De Novo Binder Design, Binding Triage, AffiniBind, Ensemble ModelingEpitope selection is the first and most consequential decision in a binder design campaign. The ideal epitope is: (1) solvent-accessible, meaning buried residues can't be contacted; (2) structurally well-ordered, since disordered loops change conformation and reduce binding predictability; (3) functionally relevant, so binding should modulate the target's activity, not just decorate its surface; and (4) unique, as the epitope should not appear on off-target proteins. AffiniBind's target analysis module maps solvent-accessible surface area, pLDDT confidence, and sequence conservation across the entire target surface to identify candidate epitopes before any binder design begins.
See also: Binding Hotspot, Interface Design, Binding Affinity (K_D)Binding hotspots are the key residues that dominate the binding free energy of a protein–protein interface. They are typically characterised by: (1) high solvent protection upon binding (large ΔSASA), (2) enrichment in aromatic and polar residues (Trp, Tyr, Arg), and (3) structural pre-organization (low conformational entropy cost). Experimental alanine scanning can identify hotspots but is labor-intensive; AffiniBind uses computational hotspot prediction to rank interface residues by their predicted contribution to binding, guiding refinement of initial designs toward the highest-value contacts.
See also: Epitope, Interface Design, dSASA (Delta Solvent-Accessible Surface Area)Related reading: Computational Strategies for Binding Hotspot IdentificationInterface residues are the subset of a protein's surface residues that directly contact the binding partner. They are computationally identified from the predicted complex structure by distance cutoffs (typically 5 Å for non-hydrogen atoms or 8 Å for Cβ atoms). AffiniBind reports the number, identity, and properties of interface residues, including hydrophobicity, secondary structure context, and predicted sidechain rotamer states, so engineers can assess whether the designed interface has the right chemical character for the intended application.
See also: Interface Contacts, Binding Hotspot, EpitopeSecondary structure composition affects both the stability and binding mode of a designed binder. Helical interfaces tend to be more rigid and predictable (lower conformational entropy penalty on binding), while loop-dominated interfaces offer greater shape adaptability at the cost of reduced binding affinity. AffiniBind reports the secondary structure composition of both the binder and the interface (helix %, sheet %, loop % in the interface region), calculated from DSSP assignment on the predicted complex structure.
See also: Interface Design, Scaffold ProteinRMSF quantifies local structural flexibility by measuring the average displacement of each residue across a structural ensemble (multiple AlphaFold model seeds, MD simulation frames, or NMR models). Low RMSF values indicate rigid, well-ordered regions (typical of hydrophobic cores and stable secondary structure elements); high RMSF values flag flexible loops, termini, and potentially disordered segments. In binder design, high RMSF at interface positions is a warning sign: a flexible interface is harder to predict accurately and incurs a larger conformational entropy penalty on binding. AffiniBind incorporates per-residue RMSF into its interface quality assessment, flagging interface positions with RMSF above 2 Å for closer inspection.
See also: pLDDT, Conformational Entropy, Secondary Structure, DSSP (Define Secondary Structure of Proteins)DSSP (also known as the Kabsch–Sander algorithm) assigns secondary structure labels to protein residues based on backbone hydrogen bonding patterns and geometric criteria. The standard output uses single-letter codes: H (α-helix), E (β-strand), and C/L (coil/loop). AffiniBind runs DSSP on every predicted complex structure to annotate interface residues by their secondary structure context, which feeds into the conformational entropy estimate and interface quality assessment. A helical interface is generally preferred over a loop-dominated one because helices are more rigid and their binding geometry is predicted with higher confidence.
See also: Secondary Structure, Interface Design, RMSF (Root Mean Square Fluctuation)AffiniBind is a vertically integrated computational pipeline (v0.5.0) that takes a target protein structure and returns a ranked shortlist of developability-validated binder candidates. The platform runs seven stages: target preparation, surface fingerprinting with four complementary deep-learning models (ScanNet, MPBind, MaSIF, APBS), binding patch identification, parallel binder generation via three cloud-accelerated engines (FreeBindCraft, PXDesign, BoltzGen), developability assessment across twelve biophysical and risk dimensions (solubility, 3D-SAP, PROPKA, PRODIGY, SSIPe, and seven additional screens), unified ranking across 24+ metrics, and multi-format results export. The pipeline is orchestrated by Nextflow with Docker-containerized execution, uses Parquet data contracts for all inter-stage communication, and provisions cloud GPU pods on-demand for generation workloads.
See also: De Novo Binder Design, Developability, Binding Triage, Composite Robustness ScoreRelated reading: Benchmarking Binding-Site Fingerprinting — DB5.5 Validation of the AffiniBind PipelineBinding triage is the most critical gate in the AffiniBind pipeline. Thousands of designs may pass the initial structural prediction step, but testing even 100 candidates experimentally is expensive and time-consuming. The triage module applies a series of increasingly stringent filters: (1) minimum pLDDT at interface positions, (2) minimum positive-state interface metrics, (3) maximum cross-reactivity against negative states, (4) developability composite score, and (5) sequence redundancy filter to remove near-duplicates. Survivors are ranked by the composite score and advanced to experimental validation.
See also: AffiniBind, Multi-State Validation (MSV), DevelopabilityThe AffiniBind unified ranking score combines 24+ raw metrics from the developability pipeline into a single advancement score. Key weighted metrics include interface confidence (fraction of low-confidence residues, buried unsatisfied polar count), binding affinity (PRODIGY ΔG), structural confidence (mean interface pLDDT), aggregation risk (spatial SAP score, hydrophobic patch size), solubility score, and manufacturing quality (protease motif count, packing quality, loop flexibility). Each metric is normalized using absolute reference points (not campaign-relative percentiles), so scores are comparable across campaigns. The composite score gates advancement alongside sequence-diversity filtering to ensure the experimental pool spans distinct design solutions.
See also: Developability, Binding Triage, AffiniBindEnsemble modeling is fundamental to AffiniBind's confidence assessment. Rather than running AlphaFold-Multimer once per design and trusting a single prediction, the pipeline generates multiple model seeds (typically 5–25) for each binder–target pair. Metrics like interface contact probability (the fraction of seeds in which a given contact appears) capture prediction robustness: a contact that appears in 25/25 seeds is far more trustworthy than one that appears in 3/25. The ensemble approach also exposes prediction instability; designs whose interface metrics vary wildly across seeds are flagged as unreliable regardless of their mean scores.
See also: AlphaFold Confidence, Binding Triage, Multi-State Validation (MSV), Sequence DiversityCASP (Critical Assessment of Structure Prediction) has been running since 1994 and is the gold-standard benchmark for structure prediction methods. Predictors receive amino acid sequences for proteins whose structures have been experimentally determined but not yet published; their predictions are then compared to the experimental structures by independent assessors. AlphaFold2's performance at CASP14 (2020), achieving median GDT_TS scores competitive with experimental methods, was a watershed moment for computational structural biology. The CASP benchmark provides the empirical foundation for trusting AlphaFold's self-assessment metrics (pLDDT, PAE) as reliable predictors of model quality.
See also: AlphaFold Confidence, CAMEO (Continuous Automated Model Evaluation), pLDDTCAMEO complements the biennial CASP experiment by providing continuous, automated evaluation. Prediction servers submit models weekly for targets released by structural biology consortia; CAMEO assesses them against the experimental structures once released, computing metrics including lDDT, CAD, and interface accuracy. For methods like AlphaFold-Multimer that are used in production pipelines such as AffiniBind, CAMEO provides ongoing validation that performance holds across diverse targets, not just the curated CASP set. WeaveSeq monitors CAMEO rankings to inform model selection and threshold calibration.
See also: CASP (Critical Assessment of Structure Prediction), AlphaFold ConfidenceScanNet applies geometric deep learning directly to 3-D atom coordinates to predict per-residue binding-site propensity. Trained on protein–protein interface data, it identifies surface residues likely to participate in binding interactions based on local structural features rather than sequence conservation. In AffiniBind's fingerprinting stage, ScanNet runs alongside MPBind, MaSIF, and APBS electrostatics, and its per-residue scores are merged into the AffiniBind learned scoring model that drives patch identification and generation targeting. Validated on the DB5.5 docking benchmark, the combined fingerprinting pipeline achieves a 96.4% Top-3 patch hit rate across 223 protein-protein complexes (see the full benchmark analysis on the News page).
See also: AffiniBind, MaSIF (Molecular Surface Interaction Fingerprinting), MPBindMaSIF represents protein surfaces as geometric point clouds (PLY meshes) and applies a surface-based neural network to predict interface site scores. Unlike atom-coordinate-based methods, MaSIF directly models the molecular surface (the actual interaction boundary), capturing shape and chemical features that are not apparent from the atomic representation alone. In AffiniBind, MaSIF precomputes surface meshes for the target and predicts per-vertex interface scores, which are then mapped to per-residue scores and merged into the unified binding-propensity score.
See also: AffiniBind, ScanNet, MPBind, Interface DesignMPBind leverages two protein language models (ProtT5 and ProstT5) to generate residue-level embeddings, then processes them through an equivariant graph neural network to predict per-residue binding propensity. The dual-language-model approach captures both sequence-level evolutionary information and structure-aware contextual features. In AffiniBind's fingerprinting stage, MPBind scores contribute to the consensus binding-propensity score alongside ScanNet, MaSIF, and APBS electrostatic analysis.
See also: AffiniBind, ScanNet, MaSIF (Molecular Surface Interaction Fingerprinting)APBS solves the Poisson–Boltzmann equation for the electrostatic potential surrounding a protein in an ionic solution. It assigns partial charges and radii using the AMBER99 force field, computes the potential on a 3-D grid, and maps it to the molecular surface. In AffiniBind, APBS provides per-residue electrostatic character (positive, negative, neutral), average surface potential, and net residue charge, information that complements the machine-learning-based binding predictions and informs the selection of electrostatically compatible interface regions.
See also: AffiniBind, ScanNet, Interface DesignFreeBindCraft performs MSA-less AlphaFold2 inference with custom templates and iterative optimization loops to design protein binders. Starting from identified binding patches on the target surface, it generates binder backbones through cycles of AF2 prediction, sequence design via ProteinMPNN, and structural refinement. Each patch yields multiple design variants with configurable binder lengths. FreeBindCraft's hotspot-centric, iterative approach produces designs with high shape complementarity and strong predicted binding energy. In the AffiniBind pipeline, it runs alongside PXDesign and BoltzGen on cloud GPU pods, and all engine outputs are pooled into the unified ranking layer for engine-agnostic candidate selection.
See also: AffiniBind, PXDesign, BoltzGen, De Novo Binder Design, ProteinMPNNPXDesign generates protein backbone structures conditioned on binding interface geometry. Unlike iterative optimization approaches, it directly samples backbone traces through a structure-conditioned generative process with configurable sampling breadth, temperature, and design count. This approach allows PXDesign to explore structural solutions that iterative methods may not reach. In the AffiniBind pipeline, PXDesign runs alongside FreeBindCraft and BoltzGen on cloud GPU pods, each contributing distinct design hypotheses to the candidate pool. All three engines' outputs are pooled and evaluated through the shared developability pipeline and unified ranking layer.
See also: AffiniBind, FreeBindCraft, BoltzGen, De Novo Binder DesignBoltzGen generates protein backbone traces through a denoising diffusion process on backbone coordinates. It takes a target epitope definition as input and generates candidate backbones by reversing a forward noising process with controlled temperature and trajectory depth. The diffusion approach allows BoltzGen to sample a broad distribution of structural solutions for each binding patch. In the AffiniBind pipeline, BoltzGen runs alongside FreeBindCraft and PXDesign on cloud GPU pods, and all three engines' designs are pooled into the unified ranking layer where they compete on identical metrics regardless of origin, ensuring engine-agnostic candidate selection.
See also: AffiniBind, FreeBindCraft, PXDesign, De Novo Binder DesignProteinMPNN (Message-Passing Neural Network) is the inverse folding engine used within AffiniBind's binder generation stage. After the generative engines (FreeBindCraft, PXDesign, BoltzGen) produce backbone traces, ProteinMPNN designs amino acid sequences predicted to fold into those structures. Unlike traditional Rosetta-based design, ProteinMPNN samples sequence space broadly and rapidly, enabling high-throughput generation of diverse binder candidates. Each generated sequence is then validated through structural prediction, creating a sequence diversity × structural accuracy matrix for downstream ranking.
See also: AffiniBind, De Novo Binder Design, Scaffold ProteinAlphaFold2, first described by Jumper et al. (2021) and validated at CASP14, uses a transformer-based architecture with evolutionary information from multiple sequence alignments (MSAs) to predict protein structures at accuracies competitive with experimental methods. In the AffiniBind pipeline, AlphaFold2 serves two roles: predicting target structures when experimental structures are unavailable (Stage 01: Target Definition), and, via AlphaFold-Multimer, predicting the structures of designed binder–target complexes (Stage 02: Binder Generation). The model also outputs the confidence metrics (pLDDT, PAE, pTM, i_pTM) that form the backbone of the pipeline's triage and ranking system. It is one of several structure prediction engines used in parallel to maximise design space coverage.
See also: AlphaFold Confidence, pLDDT, AffiniBindPoor solubility is one of the most common failure modes for computationally designed proteins. A design that looks perfect in silico may express into inclusion bodies (insoluble aggregates in E. coli) or precipitate during purification. AffiniBind's developability panel predicts solubility risk through several proxies: exposed hydrophobic patch area (larger patches → lower solubility), net charge at expression pH (near-neutral proteins tend to be less soluble), and aggregation propensity scores. The solubility score (reported as a 0–100 percentile rank in the unified ranking panel) is one of the ten ranking dimensions, weighted alongside structural confidence and interface quality.
See also: Developability, Aggregation, Composite Robustness Score, Expression Likelihood3D-SAP computes two aggregation metrics: intrinsic aggregation propensity (the product of an amino-acid-specific SAP matrix and per-residue relative solvent accessibility from FreeSASA) and spatial aggregation propensity (intrinsic values modulated by distance-decay contributions from neighboring residues). It also detects aggregation hotspots on the binder surface. In AffiniBind's developability assessment, high spatial SAP scores flag designs with aggregation risk, a critical failure mode for computationally designed proteins.
See also: AffiniBind, Developability, Solubility, AggregationPROPKA predicts pKa values for all ionizable residues by accounting for local electrostatic environments, hydrogen bonding, and desolvation effects. In AffiniBind, the per-residue pKa profiles are used to compute the net charge at pH 7 and the isoelectric point (pI) of each binder candidate. Charge distribution is an important developability factor: near-neutral proteins at expression pH tend to have better solubility, while extreme charge states can indicate aggregation risk.
See also: AffiniBind, Developability, SolubilityPRODIGY predicts the binding free energy (ΔG, in kcal/mol) and dissociation constant (Kd, in M) from the 3-D structure of a protein–protein complex using a statistical model based on interface properties including non-interacting surface (NIS) analysis. It does not require expensive free-energy perturbation or molecular dynamics simulations. In AffiniBind, PRODIGY ΔG is one of the highest-weighted metrics in the unified ranking score, reflecting the central importance of predicted binding strength in candidate selection.
See also: AffiniBind, Binding Affinity (K_D), dSASA (Delta Solvent-Accessible Surface Area), Composite Robustness ScoreSSIPe predicts the delta-delta-G of binding (ΔΔG_bind) for interface residue mutations by comparing the statistical interaction propensities of wild-type and mutant residue pairs. A negative ΔΔG indicates the mutation is predicted to strengthen binding; positive ΔΔG indicates weakening. In AffiniBind, SSIPe is applied after candidate generation to assess the binding robustness of interface mutations and identify positions where redesign could improve affinity or specificity.
See also: AffiniBind, Binding Affinity (K_D), PRODIGY (PROtein binDIng enerGY prediction)Surface Plasmon Resonance is the primary method for experimentally validating the binding kinetics of designed protein binders. The target is immobilized on a sensor chip, the binder flows over it in solution, and changes in refractive index at the surface are measured in real time to extract association rate (k_on), dissociation rate (k_off), and equilibrium affinity (K_D). AffiniBind's rank ordering is designed to correlate with SPR outcomes: designs in the top decile by composite score should show measurable binding by SPR; those in the top percentile should show specific, high-affinity binding.
See also: Binding Affinity (K_D), BLI (Bio-Layer Interferometry)Related reading: High-Throughput Validation of Synthetic BindersBio-Layer Interferometry uses fibre-optic biosensors to measure binding in real time. The binder is loaded onto the biosensor tip, which is then dipped into a solution containing the target. Binding causes a shift in the interference pattern of white light reflected from two surfaces on the tip, proportional to the mass of bound material. BLI is particularly useful for screening in complex backgrounds (e.g., serum or cell lysate) where SPR's microfluidic format can foul, making it a natural fit for the complex-serum validation step in WeaveSeq's Development Pathway.
See also: SPR (Surface Plasmon Resonance), Binding Affinity (K_D), Complex Serum ValidationRelated reading: High-Throughput Validation of Synthetic BindersCell-free expression uses E. coli, wheat germ, or rabbit reticulocyte lysates to transcribe and translate DNA into protein in a test tube. In the AffiniBind workflow, IVTT is the Stage 3 screening step: hundreds of binder candidates are expressed in parallel, and the crude lysate is screened for target binding via BLI or ELISA. Successful candidates advance to recombinant production (Stage 4). IVTT dramatically accelerates the design → screen cycle by eliminating the need for cell transformation, colony picking, and individual culture growth.
See also: BLI (Bio-Layer Interferometry), Development PathwayThe Development Pathway is WeaveSeq's operational framework for turning a target into a validated binder. The six stages are: (1) Feasibility Review: computational target assessment, epitope identification, and Go/No-Go recommendation; (2) Binder Discovery: de novo design and AffiniBind triage to a ranked shortlist; (3) IVTT Screening: cell-free expression and binding verification of top candidates; (4) Recombinant Production: scalable expression and purification of lead candidates; (5) Functional Validation: SPR/BLI kinetics and functional assay testing; (6) Assay Development: integration of the validated binder into the end-user's assay format. At each gate, the client receives a data package and decides whether to continue, pause, or redirect.
See also: AffiniBind, Binding Triage, IVTT (In Vitro Transcription–Translation)A binder that performs perfectly in clean buffer may fail completely in serum; non-specific binding to abundant serum proteins (albumin, immunoglobulins) can mask specific target engagement. Complex serum validation is a development-stage assay that evaluates binder performance in biologically relevant backgrounds: the target is spiked into serum at known concentrations, and the binder's ability to capture it is measured via BLI or pull-down. This validation step bridges the gap between computational specificity predictions and real-world assay conditions. In WeaveSeq's Development Pathway, complex serum validation occurs during Stage 5 (Functional Validation) as part of matrix compatibility testing.
See also: BLI (Bio-Layer Interferometry), Multi-State Validation (MSV), Specificity, Development Pathway, ELISA (Enzyme-Linked Immunosorbent Assay)ELISA is a foundational assay format in protein engineering and validation. In a typical sandwich ELISA, a capture antibody (or designed binder) is immobilised on a plate, the target is added and captured, and a detection antibody produces an enzymatic colour change proportional to target concentration. Designed protein binders can replace traditional antibodies at either the capture or detection position, or both. WeaveSeq's Assay Development stage (Stage 6) delivers binders optimised for ELISA compatibility, including matched capture/detection pairs, signal-to-noise characterisation, and matrix effect assessment.
See also: SPR (Surface Plasmon Resonance), BLI (Bio-Layer Interferometry), Development Pathway, IVTT (In Vitro Transcription–Translation), Complex Serum ValidationA computationally perfect binder that expresses at 0.1 mg/L and precipitates during buffer exchange is not a product; it's a research curiosity. Manufacturability assessment evaluates practical production factors: expression titre (mg of purified protein per litre of culture), purification yield and purity (single band on SDS-PAGE, low endotoxin), concentration stability (no aggregation at >1 mg/mL), freeze–thaw robustness, and long-term storage behaviour. AffiniBind's ranking incorporates manufacturability predictions (expression likelihood, solubility, aggregation risk), but the ultimate assessment comes from recombinant production data (Stage 4) and functional validation (Stage 5).
See also: Developability, Composite Robustness Score, Expression Likelihood, SolubilityAggregation is the #1 reason computationally designed proteins fail in the wet lab. A design with perfect interface metrics is useless if it forms inclusion bodies in E. coli or precipitates in PBS. AffiniBind's developability panel screens for aggregation risk through: exposed hydrophobic patch detection (patches larger than 400 Ų are flagged), oligomerization propensity prediction, self-binding risk assessment, and charge distribution analysis. Designs with multiple aggregated risk flags are deprioritised regardless of their binding scores.
See also: Developability, Composite Robustness Score, SolubilityWhen a flexible protein or loop binds a target, it loses conformational freedom. This entropy penalty must be overcome by favourable binding enthalpy (hydrogen bonds, van der Waals contacts, hydrophobic burial). Pre-organized, rigid scaffolds minimise this penalty, which is why AffiniBind favours helical and sheet-rich binders over flexible loop-based ones. The platform estimates conformational entropy indirectly through secondary structure composition and per-residue RMSF (root mean square fluctuation) from the AlphaFold prediction, flagging designs with excessive predicted flexibility in the interface region.
See also: Interface Design, Secondary Structure, Binding Affinity (K_D)The hydrophobic effect is not a true 'force'; it arises because water molecules near non-polar surfaces are entropically constrained (they form ordered cages, or clathrates). Burying non-polar surface in a protein–protein interface releases this constrained water, generating a favourable entropic contribution to binding. This is why interfaces are enriched in hydrophobic residues relative to the protein surface, and why dSASA (which captures the burial of non-polar surface area) correlates with binding affinity. However, an interface that is too hydrophobic will aggregate. The design must balance burial of non-polar surface with a polar 'rim' that maintains solubility.
See also: dSASA (Delta Solvent-Accessible Surface Area), Binding Affinity (K_D), AggregationPoor solubility is one of the most common failure modes for computationally designed proteins. A design that looks perfect in silico may express into inclusion bodies (insoluble aggregates in E. coli) or precipitate during purification. AffiniBind's developability panel predicts solubility risk through several proxies: exposed hydrophobic patch area (larger patches → lower solubility), net charge at expression pH (near-neutral proteins tend to be less soluble), and aggregation propensity scores. The solubility score (reported as a 0–100 percentile rank in the unified ranking panel) is one of the ten ranking dimensions, weighted alongside structural confidence and interface quality.
See also: Developability, Aggregation, Composite Robustness Score, Expression LikelihoodExpression likelihood translates sequence features into a practical forecast: will this design produce protein when we put the DNA into E. coli? The score incorporates codon adaptation index (CAI) relative to the expression host, predicted mRNA secondary structure at the translation initiation site, GC content, and sequence-based solubility and expression models trained on large-scale expression datasets. A high expression likelihood does not guarantee success, but a low score reliably predicts failure; designs in the bottom quartile rarely produce usable protein. AffiniBind's ranking weights expression likelihood alongside solubility and aggregation risk to ensure shortlisted candidates have a realistic chance of surviving the IVTT and recombinant production stages.
See also: Solubility, Developability, Composite Robustness Score, IVTT (In Vitro Transcription–Translation), ManufacturabilityAssay-aware design is WeaveSeq's core engineering philosophy. Traditional computational design optimizes for predicted binding affinity alone, but a binder that works perfectly in silico may fail in an ELISA because of surface adsorption, matrix interference, or poor expression yield. Assay-aware design means we evaluate every candidate against the constraints of the intended assay format from the start: the surface chemistry of the plate or chip, the signal generation mechanism, the sample matrix (buffer, serum, lysate), and the practical requirements of manufacturing and storage. This shifts the question from 'does it bind?' to 'will it work in your assay?'
See also: Developability, Manufacturability, Development Pathway, SpecificityNextflow DSL2 is the orchestration layer that coordinates the AffiniBind computational pipeline. It manages the execution order of the seven pipeline stages, provisions cloud resources on demand, handles artifact exchange between stages via S3, and caches intermediate results so only changed inputs trigger re-computation. Each computational tool (AlphaFold, ScanNet, MaSIF, FreeBindCraft, etc.) runs in its own Docker container, ensuring reproducibility and eliminating environment-dependent failures.
See also: AffiniBindApache Parquet is a columnar data format designed for efficient analytics and data interchange. In the AffiniBind pipeline, every stage writes a Parquet file with a canonical PyArrow schema that the next stage consumes. This ensures: type-safe data flow (no silent type coercion), complete lineage tracking (every computed value can be traced to its source), schema validation at stage boundaries (failures produce diagnostics, not empty columns), and incremental re-execution (the orchestration layer detects which inputs have changed and only reruns affected stages).
See also: AffiniBind, Nextflow (DSL2)Understanding the metrics is step one. Bring us your target and we will apply them — from epitope mapping through candidate ranking — to deliver assay-ready binders.