# US 2026/0094666 A1 · Fold conditioned protein structure generation

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0094666 A1
- Document ID: US20260094666A1
- Status: Published application
- Kind: A1
- Application number: 19/234,113
- Filed: 2025-06-10
- Published: 2026-04-02
- Earliest claimed: 2024-10-01
- Assignees: NVIDIA Corporation
- Applicants: NVIDIA Corporation
- Inventors: Karsten Kreis; Tomas Geffner; Kieran Didi; Zuobai Zhang; Arash Vahdat; Danny Reidenbach; Zhonglin Cao; Emine Kucukbenli; Mario Geiger; Chris Dallago
- Web page: https://patentgrep.com/patent/US20260094666A1

## CPC

- G16B 15/20: Protein or domain folding
- G06N 3/045: Combinations of networks
- G06N 3/0475: Generative networks
- G16B 40/20: Supervised data analysis
- G16B 40/30: Unsupervised data analysis
- G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

De novo protein design, the rational design of new proteins from scratch with specific functions and properties, is a grand challenge in molecular biology. Recently, deep generative models have emerged as a novel data-driven tool for protein engineering. However, current diffusion- and flow-based models generally synthesize backbones only, without sequence or side chains, while protein language models often model sequences instead. The present disclosure provides flow-based protein structure generation which can be conditioned on a given fold class.

## Claims

### Claim 1

A method, comprising:
at a device:
generating, by a flow-based generative model conditioned on an input fold class parameter indicating one or more fold classes, a synthetic protein structure of the one or more fold classes; and
outputting the synthetic protein structure.

### Claim 2 (depends on 1)

The method of claim 1, wherein the input fold class parameter indicates a single fold class.

### Claim 3 (depends on 2)

The method of claim 2, wherein the flow-based generative model generates the synthetic protein structure of the single fold class.

### Claim 4 (depends on 1)

The method of claim 1, wherein the input fold class parameter is hierarchical.

### Claim 5 (depends on 4)

The method of claim 4, wherein the input fold class parameter indicates a primary fold class and a secondary fold class.

### Claim 6 (depends on 5)

The method of claim 5, wherein the input fold class parameter indicates a first degree to which the synthetic protein structure is to be generated in accordance with the primary fold class and a second degree to which the synthetic protein structure is to be generated in accordance with the secondary fold class.

### Claim 7 (depends on 1)

The method of claim 1, wherein the synthetic protein structure is a synthetic protein backbone.

### Claim 8 (depends on 1)

The method of claim 1, wherein the flow-based generative model generates the synthetic protein structure by:
creating a sequence representation,
creating sequence conditioning features,
creating a pair representation, and
processing the sequence representation, the sequence conditioning features, and the pair representation by a neural network comprised of multi-head attention layers to generate the synthetic protein structure.

### Claim 9 (depends on 8)

The method of claim 8, wherein the sequence representation and the pair representation are created from noisy protein coordinates.

### Claim 10 (depends on 9)

The method of claim 9, wherein the sequence representation is created to include registers that capture global information.

### Claim 11 (depends on 8)

The method of claim 8, wherein the sequence conditioning features are created from the input fold class parameter.

### Claim 12 (depends on 8)

The method of claim 8, wherein the multi-head attention layers are conditioned on the sequence conditioning features and are biased through the pair representation to update the sequence representation.

### Claim 13 (depends on 8)

The method of claim 8, wherein the sequence representation, the sequence conditioning features, and the pair representation are normalized prior to processing through the multi-head attention layers.

### Claim 14 (depends on 12)

The method of claim 12, wherein the neural network is configured to update the pair representation based on the updated sequence representation and to decode the updated pair representation into pairwise distances for the updated sequence representation.

### Claim 15 (depends on 14)

The method of claim 14, wherein the neural network is comprised of triangle multiplicative layers for updating the pair representation.

### Claim 16 (depends on 1)

The method of claim 1, wherein classifier-free guidance is used to condition the flow-based generative model on the input fold class parameter.

### Claim 17 (depends on 1)

The method of claim 1, wherein autoguidance is used to guide generation of the synthetic protein structure by the flow-based generative model.

### Claim 18 (depends on 1)

The method of claim 1, wherein the flow-based generative model is trained on training data comprised of sample protein structures labeled with fold class labels.

### Claim 19 (depends on 18)

The method of claim 18, wherein the fold class labels are hierarchical to indicate one or more fold classes for each sample protein structure.

### Claim 20 (depends on 18)

The method of claim 18, wherein the flow-based generative model is trained in at least two training stages each using a different set of training data.

### Claim 21 (depends on 20)

The method of claim 20, wherein the at least two training stages include:
a first training stage in which the flow-based generative model is trained on a first set of sample protein structures having a sequence length below a defined threshold, and
a second training stage in which the flow-based generative model is trained on a second set of sample protein structures having a sequence length above the defined threshold.

### Claim 22

A system, comprising:
a non-transitory memory storage comprising instructions; and
one or more processors in communication with the memory, wherein the one or more processors execute the instructions to:
generate, by a flow-based generative model conditioned on an input fold class parameter indicating one or more fold classes, a synthetic protein structure of the one or more fold classes; and
output the synthetic protein structure.

### Claim 23 (depends on 22)

The system of claim 22, wherein the input fold class parameter indicates a first degree to which the synthetic protein structure is to be generated in accordance with a primary fold class and a second degree to which the synthetic protein structure is to be generated in accordance with a secondary fold class.

### Claim 24

A non-transitory computer-readable media storing computer instructions which when executed by one or more processors of a device cause the device to:
generate, by a flow-based generative model conditioned on an input fold class parameter indicating one or more fold classes, a synthetic protein structure of the one or more fold classes; and
output the synthetic protein structure.

### Claim 25 (depends on 24)

The non-transitory computer-readable media of claim 24, wherein the input fold class parameter indicates a first degree to which the synthetic protein structure is to be generated in accordance with a primary fold class and a second degree to which the synthetic protein structure is to be generated in accordance with a secondary fold class.
