---
title: 'HelixFold-S1: MSA-free & Contact-Guided Models'
url: https://www.emergentmind.com/topics/helixfold-s1
type: topic
---

# HelixFold-S1: MSA-free & Contact-Guided Models

HelixFold-S1 is a name used for two distinct models within the HelixFold line. In the 2022 literature and PaddleHelix ecosystem, “HelixFold-S1” commonly denotes **HelixFold-Single**, an end-to-end, **MSA-free protein structure predictor** that replaces AlphaFold2’s MSA stack with a large protein language model while retaining most of AlphaFold2’s geometric machinery [2207.13921]. In the 2025 literature, HelixFold-S1 denotes a **sampling-optimized variant of HelixFold3** for **biomolecular complex structure prediction**, introducing explicit intermolecular contact prediction and contact-guided test-time sampling within an AlphaFold3-like architecture [2507.09251]. The shared label therefore refers either to a single-sequence monomer-folding model or to a contact-guided complex-sampling model, depending on context.

## 1. Nomenclature and lineage

The ambiguity of the term reflects the evolution of the HelixFold family. “HelixFold” originally referred to an efficient PaddlePaddle implementation of AlphaFold2, with systems-level optimization for training and inference speed and memory reduction, while maintaining accuracy on CASP14 and CAMEO comparable to AlphaFold2 [2207.05477]. On top of that base, HelixFold-Single was introduced as a single-sequence predictor and was often abbreviated as “HelixFold-S1” in code and deployments [2207.13921]. Later, a different model, also named HelixFold-S1, was introduced as a sampling-optimized HelixFold3 variant for complex prediction [2507.09251].

| Usage of “HelixFold-S1” | Base model | Primary task |
|---|---|---|
| HelixFold-Single | HelixFold / AlphaFold2-style geometry | Single-sequence protein structure prediction |
| HelixFold-S1 | HelixFold3 / AlphaFold3-like architecture | Biomolecular complex structure prediction via contact-guided sampling |

The 2022 and 2025 usages are therefore not minor revisions of a single model family member. They occupy different positions in the HelixFold trajectory: the earlier system addresses the removal of MSA dependence for protein folding, whereas the later system addresses inference-time exploration of docking modes for complexes when co-evolutionary signals are weak or absent [2207.13921] [2507.09251].

A plausible implication is that references to “HelixFold-S1” in software, deployments, or secondary discussions require disambiguation by year, task, or architectural context. Without that disambiguation, the label conflates two substantially different modeling objectives.

## 2. HelixFold-Single as the 2022 meaning of HelixFold-S1

HelixFold-Single is an end-to-end, MSA-free protein structure predictor designed to infer 3D structures directly from a **single amino-acid sequence**, without computing MSAs or using templates [2207.13921]. Its stated motivation is that MSA search is time-consuming, often taking dozens of minutes per protein, and that many downstream workloads require large numbers of predictions, making MSA-based inference a bottleneck [2207.13921].

The model substitutes AlphaFold2’s explicit evolutionary input pathway with a **large protein language model (PLM)**. The PLM is trained on primary sequences by masked language modeling and is used as an alternative mechanism for learning co-evolutionary information [2207.13921]. The resulting architecture keeps the essential geometric learning stack of AlphaFold2—specifically a modified Evoformer and Structure Module—but removes the MSA stack and template search at inference [2207.13921].

The pipeline has three major components. First, the **PLM Base** takes the primary sequence \( \mathbf{x} = (x_1,\dots,x_L) \) and produces per-residue embeddings and per-layer multi-head attention maps [2207.13921]. Second, an **Adaptor** linearly projects those outputs into the dimensions required by the folding trunk, producing an initial single representation and an initial pair representation:
\[
\begin{aligned}
\tilde{\mathbf{x}}^{(0)} &= \text{Linear}\bigl(\mathbf{x}^{(n_{\text{PLM}})}\bigr), \\
\tilde{\mathbf{z}}^{(0)} &= \text{Linear}\Bigl([\mathbf{z}^{(1)}, \mathbf{z}^{(2)}, \dots, \mathbf{z}^{(n_{\text{PLM}})}]\Bigr).
\end{aligned}
\]
Third, **Geometric Modeling** applies a revised Evoformer and a Structure Module, with recycling, to predict atomic coordinates [2207.13921].

Within the PLM, the paper uses a DeBERTa-based Transformer with disentangled attention and switches from Post-Norm to Pre-Norm for stability [2207.13921]. For the main model, the **PLM-1B** configuration has approximately **1.09B parameters**, \( n_{\text{PLM}} = 20 \), hidden size \( d_{\text{PLM}} = 2048 \), intermediate size 8192, and \( h_{\text{PLM}} = 16 \) attention heads [2207.13921]. The Evoformer has approximately **87M parameters**, and the Structure Module approximately **1.7M parameters**, for a total of about **1.18B parameters** plus adaptor layers [2207.13921].

A central architectural detail is that the pair representation is derived from attention maps across all PLM layers, while the single representation comes from the final-layer residue embeddings [2207.13921]. This makes the model an explicit attempt to relocate MSA-like statistical structure into the PLM, then reuse AlphaFold2-level geometric reasoning downstream.

## 3. Architecture, training regime, and empirical properties of HelixFold-Single

HelixFold-Single’s revised Evoformer closely follows AlphaFold2 but removes **column-wise gated self-attention**, since there is no MSA dimension to attend over in the MSA-free setting [2207.13921]. The input dimensions after the adaptor are \( \tilde{\mathbf{x}}^{(0)} \in \mathbb{R}^{L \times 512} \) and \( \tilde{\mathbf{z}}^{(0)} \in \mathbb{R}^{L \times L \times 64} \) [2207.13921]. The Structure Module follows AlphaFold2 and uses **Invariant Point Attention (IPA)** and SE(3)-equivariant operations to infer residue frames, torsions, and atomic coordinates [2207.13921].

Training proceeds in two stages [2207.13921]. In **Stage 1**, the PLM is pre-trained on **UniRef30 (2021-03)**, comprising about **260–300 million sequences**, using masked language modeling with 15% residue masking [2207.13921]. In **Stage 2**, the PLM Base and Geometric Modeling are **jointly optimized** with structural supervision on about **1.1M structures**, combining filtered RCSB PDB structures with two distillation datasets derived from AlphaFold2 and the AlphaFold Protein Structure Database [2207.13921]. The structure prediction objective follows AlphaFold2, including **FAPE** and auxiliary losses such as distogram, angle/torsion, violation, and pLDDT-related terms, although the paper does not rederive all formulas [2207.13921].

The main reported runtime result is that inference avoids external database search entirely. On a single NVIDIA A100 (40G), median inference times for HelixFold-Single are reported as **1.5 s** for proteins of length 1–100 and 100–200, **2.1 s** for 200–400, **6.2 s** for 400–800, and **37.5 s** for \(>800\), compared with AlphaFold2 total pipeline times of **766.1 s**, **795.8 s**, **908.3 s**, **1125.2 s**, and **1611.2 s**, respectively [2207.13921]. The paper states that for sequences shorter than 100 residues, HelixFold-Single is roughly **1000× faster** than the full AlphaFold2 pipeline, while for \(>800\) residues it remains approximately **40–50× faster** [2207.13921].

On **CASP14**, HelixFold-Single significantly outperforms the MSA-free baselines discussed in the paper but still lags MSA-based AlphaFold2 overall, especially on challenging FM targets [2207.13921]. On **CAMEO**, its average accuracy is described as comparable to AlphaFold2 with MSA input and better than RoseTTAFold with MSA input [2207.13921]. When stratified by MSA depth, the reported behavior is that HelixFold-Single improves as MSA depth increases, and for targets with **MSA depth \(>1000\)** its TM-scores are comparable to MSA-based methods [2207.13921].

The paper also reports negative correlations between PLM **perplexity** and both MSA depth and downstream TM-score [2207.13921]. This suggests that PLM modeling quality functions as a proxy for how much family-level information the model has internalized. The authors further show that scaling the PLM from **100M** to **1B** parameters lowers perplexity and improves long-range contact prediction measured by **top \(L/5\) long-range contact precision \(P@L/5\)** [2207.13921].

## 4. HelixFold-S1 as the 2025 contact-guided sampling model

The 2025 HelixFold-S1 addresses a different problem: **biomolecular complex structure prediction** across interaction types including **protein-antibody, protein-protein, protein-ligand, protein-RNA, and protein-DNA** [2507.09251]. Its stated motivation is that co-evolutionary signals are often weak or absent for these tasks—especially for antibody–antigen, protein–ligand, and many nucleic-acid complexes—and that pure large-scale sampling can yield diminishing returns because many extra samples remain minor variants of the same binding mode [2507.09251].

HelixFold-S1 is defined as a **sampling-optimized variant of HelixFold3** that explicitly predicts intermolecular contacts and uses them to guide test-time sampling [2507.09251]. Conceptually, it converts “aimless” sampling into **targeted**, contact-driven exploration of docking modes while remaining within an AlphaFold3-like architecture [2507.09251].

The model introduces two contact-centric additions [2507.09251]. The **Contact Prediction Module** takes pair representations from HF3’s Pairformer and outputs contact probabilities between tokens from different chains. The contact definition is binary: tokens \(i\) and \(j\) are in contact if **any atom pair** across the two tokens is within **5 Å** [2507.09251]. The **Contact Conditioning Module** accepts a binary matrix \( \{c_{ij}\} \), with \( c_{ij} \in \{0,1\} \), projects it into pair-feature space, and fuses it into the pair representation at the **Input Embedder** stage so that the diffusion model is conditioned on specified contacts [2507.09251].

The 2025 paper describes a **three-stage inference pipeline** [2507.09251]. In the first stage, the model is run in contact-prediction mode **five times** by default and the resulting contact probability matrices are averaged, with intra-chain entries masked out [2507.09251]. In the second stage, inter-chain token pairs are sorted by predicted contact probability and iteratively selected as constraints, using **redundant contact pruning** so that contacts already realized in previous predictions are not repeatedly used for new samples [2507.09251]. When the maximum remaining contact probability falls below
\[
\text{threshold} = 0.2 \cdot p_{\max},
\]
where \( p_{\max} = \max_{i,j} \bar{p}_{ij}^{\text{contact}} \), the system switches from exploring new contacts to **enriched sampling** from previously realized contact sets until the sample budget is exhausted [2507.09251].

In the third stage, the selected contact or contact subset is converted into a binary contact matrix and passed through the Contact Conditioning Module to generate a full 3D structure [2507.09251]. Reported experiments use **10 recycles** and **200 diffusion steps** per sample [2507.09251]. Candidate structures are ranked by
\[
\text{ranking\_confidence} = 0.2 \cdot \text{pTM} + 0.8 \cdot \text{ipTM} - 1.0 \cdot \text{has\_clash},
\]
and the highest-confidence structure is returned as the final prediction [2507.09251].

The paper emphasizes the exploration–exploitation structure of the sampler: early high-probability contact proposals broaden the search space, while later enriched sampling refines already promising interface patterns [2507.09251]. A plausible implication is that the contact map acts as an intermediate, lower-dimensional interface representation that makes diversity controllable rather than incidental.

## 5. Training, evaluation, and observed behavior of the 2025 HelixFold-S1

The 2025 HelixFold-S1 retains HelixFold3’s multi-term loss and adds an inter-chain binary contact loss [2507.09251]. The total loss is given as
\[
\mathcal{L}_{\text{loss}} =
\alpha_{\text{confidence}} \mathcal{L}_{\text{confidence}} +
\alpha_{\text{diffusion}} \mathcal{L}_{\text{diffusion}} +
\alpha_{\text{distogram}} \mathcal{L}_{\text{distogram}} +
\alpha_{\text{contact}} \mathcal{L}_{\text{contact}},
\]
with \( \alpha_{\text{confidence}} = 0.01 \), \( \alpha_{\text{diffusion}} = 4 \), \( \alpha_{\text{distogram}} = 0.3 \), and \( \alpha_{\text{contact}} = 1 \) only for contact-prediction samples and 0 for contact-conditioned structure-prediction samples [2507.09251]. The contact loss is binary cross-entropy over inter-chain token pairs [2507.09251].

Training is organized into **three fine-tuning stages** [2507.09251]. Stage 1 trains only contact-conditioned structure prediction, with **70%** probability of using **1–10** true inter-chain contacts as constraints and **30%** probability of using none; the crop size is **384**, the training length **10,000 steps**, and the optimizer is Adam with \( \beta_1 = 0.9 \), \( \beta_2 = 0.95 \), \( \epsilon = 10^{-8} \), learning rate \( 2 \times 10^{-4} \), and batch size **128** [2507.09251]. Stage 2 jointly trains contact prediction and contact-conditioned structure prediction for **20,000 steps** at crop size **384**, and Stage 3 repeats that setup with crop size **640** for **3,000 steps** [2507.09251]. The new contact modules are randomly initialized, while all other parameters are copied from HF3 [2507.09251].

Evaluation sets are drawn from PDB and SAbDab in **2022–2024**, filtered to be non-redundant and test-only relative to HF3’s training set [2507.09251]. The benchmark includes **221 protein–antibody complexes** from SAbDab, **199 protein–protein**, **194 protein–ligand**, **94 protein–RNA**, and **252 protein–DNA** complexes [2507.09251]. Reported metrics are **DockQ** for protein–protein and antibody–antigen, **interface lDDT (iLDDT)** for protein–nucleic-acid complexes, and symmetry-corrected ligand heavy-atom **RMSD** after pocket alignment for protein–ligand complexes, with the main ligand metric being the fraction of predictions with ligand RMSD \(< 2\) Å [2507.09251].

The baseline set includes unguided **HF3**, **HF3 w/ AFsample**, **Chai-1**, **Boltz-2**, and **AFsample** itself for sampling-strategy comparison [2507.09251]. Across complex types, both HF3 and HF-S1 improve as the number of samples increases, but under **oracle selection** the performance gap between HF-S1 and HF3 widens with more sampling, indicating that HF-S1 discovers better structures within its ensemble because its search is more diverse and better targeted [2507.09251]. The paper highlights especially large gains on **protein–antibody**, **protein–ligand**, and **moderately difficult nucleic-acid** complexes [2507.09251].

For a **2024-only protein–antibody subset** of **95 complexes** using **80 samples**, with success defined as **DockQ \(> 0.23\)**, the paper reports that HF3 and Boltz-2 are strong baselines but show limited improvement with additional samples, Chai-1 shows virtually no gain from additional sampling, HF3 w/ AFsample improves with more samples, and HF-S1 shows substantially faster and larger performance improvements than all baselines [2507.09251]. In contact-focused analyses, the predicted contact matrix of HF-S1 has substantially higher **AUPRC** than a posterior contact matrix derived from HF3 samples, protein–antibody complexes have the lowest AUPRC, and protein–ligand complexes the highest [2507.09251]. The **contact satisfaction rate** is reported as **\(>70\%\)** across complex types, indicating that conditioning usually yields structures satisfying the imposed contacts [2507.09251].

The paper also analyzes target-level contact probability \( p_{\max} \) as an empirical difficulty indicator [2507.09251]. Lower \( p_{\max} \) is associated with lower single-sample accuracy in DockQ, iLDDT, and ligand RMSD, whereas higher \( p_{\max} \) is associated with stronger single-sample predictions [2507.09251]. Intermediate \( p_{\max} \) values show the largest gains under oracle selection, suggesting that these are the cases where contact-guided sampling is most beneficial [2507.09251]. The authors stop short of implementing adaptive sampling allocation, but the reported analysis supports that interpretation [2507.09251].

## 6. Relation to the broader HelixFold ecosystem and interpretive cautions

The broader HelixFold program begins with **HelixFold**, a PaddlePaddle implementation of AlphaFold2 optimized by **operator fusion**, **tensor fusion**, **hybrid parallelism**, **Recompute**, **BFloat16**, and in-place memory read/write [2207.05477]. HelixFold reports end-to-end training in **7.5 days** on 128 A100 GPUs and **5.3 days** with hybrid parallelism, compared with about **11 days** for the original AlphaFold2 and OpenFold, while maintaining near-parity on CASP14 and CAMEO [2207.05477]. That work does not define HelixFold-S1 explicitly, but it provides the implementation substrate on which later variants were built [2207.05477].

Against that backdrop, the two meanings of HelixFold-S1 exemplify two different research directions. The 2022 HelixFold-Single asks whether the MSA stack can be replaced by a large PLM while retaining AlphaFold2’s geometric trunk [2207.13921]. The 2025 HelixFold-S1 asks whether complex prediction can be improved at inference time by replacing indiscriminate sampling with contact-guided exploration and contact-conditioned generation [2507.09251]. The former is primarily an **input-representation substitution** for monomer folding; the latter is primarily an **inference-time search strategy** for complexes.

Several common misconceptions follow from the name collision. One is to treat HelixFold-S1 as a single, continuously upgraded model. The source material does not support that interpretation: the 2022 and 2025 systems differ in task, architecture, and training setup [2207.13921] [2507.09251]. Another is to assume that the 2025 S1 is simply a continuation of the “Single” naming convention. The paper explicitly presents it instead as a **sampling-optimized variant of HelixFold3**, and the “S1” designation is not expanded in the text; the authors only infer from context that it denotes a sampling-oriented variant [2507.09251].

A plausible implication is that “HelixFold-S1” should be read as a context-dependent identifier rather than a stable taxonomic label. In bibliographic usage, the safest interpretation is therefore to pair the name with its arXiv record, publication year, and task domain: **HelixFold-Single** for MSA-free protein structure prediction [2207.13921], or **HelixFold-S1** for contact-guided biomolecular interaction structure prediction [2507.09251].

Source: https://www.emergentmind.com/topics/helixfold-s1