feat: steering to avoid chain breaks and clashes #171

ludwigwinkler · 2025-10-15T12:33:25Z

Add Steering functionality to BioEmu

…u into luwinkler/fk_steering

…ith FK sampling

- Enhanced the `steering.py` module with additional plotting functions for Ca-Ca distances and clashes. - Introduced a new `steering_run.py` script for testing sample generation with steering, utilizing Hydra for configuration management. - Created a scratch pad script for testing loss functions visually. - Updated the test suite in `test_steering.py` to include WandB logging and improved configuration handling with Hydra. - Removed deprecated code and organized potential classes for better clarity and maintainability.

- Introduced a new `bioemu.mdc` file containing development guidelines for the BioEMU project, covering molecular dynamics, blob storage, curated MD data structure, file paths, analysis patterns, and error handling. - Added a `load_md.py` script demonstrating how to interact with blob storage and load molecular dynamics data, including trajectory analysis using MDTraj. - Updated the `run_steering_comparison.py` script to iterate over different particle counts for steering experiments, improving configurability and analysis capabilities. - Enhanced the `denoiser.py` with a new `euler_maruyama_denoiser` function and integrated it into the testing framework. - Updated configuration files for steering and denoising to reflect new parameters and functionalities.

…odule - Updated the `bioemu.mdc` file to provide a comprehensive development guide, including architectural principles, design patterns, and implementation guidelines for the BioEMU project. - Added a new `analytical_diffusion.py` module that implements a time-dependent Gaussian Mixture Model for analytical diffusion, including functionality for forward and reverse diffusion processes. - Refactored the `load_md.py` script by removing unused imports to streamline the code. - Enhanced the `run_steering_comparison.py` script to improve configurability and analysis of steering experiments, including adjustments to plotting and data handling. - Introduced a new `stratified_sampling.py` module with tests for stratified resampling functionality. - Added a `sweep_analysis.py` module for analyzing sweep data from Weights and Biases, including visualization of results. - Updated the `steering.yaml` configuration to reflect changes in potential parameters for steering functionality.

…n, and denoiser scripts - Changed import statement for tqdm from `tqdm.auto` to `tqdm` for consistency across modules. - Added plt.show() in analyze_termini_distribution function to ensure plots are displayed. - Commented out plt.show() in main function to prevent automatic display during batch processing.

…nto luwinkler/fk_steering

- Refactored steering module to include ChainBreakPotential and ChainClashPotential, replacing previous distance potentials. - Updated run_steering_comparison.py to refine steering configurations, including adjustments to num_samples and particle counts. - Implemented fast steering optimization to delay particle creation until steering start time for improved performance. - Added validation for steering configurations and assertions to ensure expected steering execution. - Introduced comprehensive tests for new steering features, including physical and fast steering capabilities. - Updated steering.yaml configuration to reflect new potential parameters and added end time for steering.

- Introduced a new section in the README for "Steering for Enhanced Physical Realism," detailing the use of Sequential Monte Carlo for guiding protein structure diffusion. - Added example commands for enabling steering via CLI and Python API, including key parameters and potential configurations. - Created a new `hydra_run.py` script for running BioEMU sampling with Hydra configuration management, allowing for easier experimentation with steering parameters. - Updated existing scripts to reflect changes in steering configuration, including renaming parameters for clarity and consistency. - Added a new `README_hydra_run.md` to document the usage of the Hydra-based entry point. - Implemented tests for CLI integration, ensuring that steering functionality works as expected through command-line parameters.

- Updated the README to clarify the steering process, including the default behavior for steering potentials and the use of multiple particles. - Removed references to the now-optional `steering_potentials_config` parameter in example commands and clarified its default behavior. - Enhanced the sample.py script to load default steering potentials when no custom configuration is provided, improving usability. - Added warnings for missing default configuration files to aid in troubleshooting.

…ctions - Changed the section title from "Steering for Enhanced Physical Realism" to "Steering structures" for clarity. - Updated CLI instructions to specify the requirement of setting `--num_steering_particles` to greater than 1 for enabling steering. - Removed Python API example for steering to streamline the documentation and focus on CLI usage.

…ydra configuration section - Added a Python API example for steering, demonstrating how to use the `bioemu.sample` module. - Removed the section detailing the Hydra configuration interface to streamline the documentation and focus on the primary usage methods.

- Added an entry to .gitignore to ignore all files in the docs directory, preventing them from being tracked by Git.

- Updated the `run_steering_comparison.py` and `run_guidance_steering_comparison.py` scripts to streamline steering configuration handling. - Introduced a new `DisulfideBridgePotential` class for guiding disulfide bridge formation, including parameters for specified cysteine pairs. - Added a new configuration file for disulfide steering and updated existing steering configurations to reflect changes in potential definitions. - Enhanced the `sample.py` module to support the new steering configuration structure, allowing for better integration of disulfide bridge steering. - Implemented tests for the `DisulfideBridgePotential` to ensure correct functionality and energy calculations.

- Introduced new guidance steering configuration and potential for enhanced structural constraints. - Updated `run_guidance_steering_comparison.py` to support three-way comparison: no steering, resampling only, and guidance steering. - Added new binary images for visualization of steering comparisons. - Refactored `run_steering_experiment` to accommodate the new experiment type parameter. - Enhanced analysis functions to compare termini distances and KL divergence across different steering methods. - Updated existing steering configurations to include guidance steering options and parameters. - Improved error handling and logging for better user feedback during experiments.

…urations - Updated `run_guidance_steering_comparison.py` to streamline handling of guidance steering alongside resampling. - Refactored steering configuration to improve clarity and functionality, including adjustments to learning rates and steps for guidance. - Modified `dpm_solver` to apply gradient guidance more effectively and ensure correct score calculations. - Increased sample size in main configuration for better statistical analysis during experiments. - Updated binary image for visualization of steering comparisons.

- Updated `disulfide_steering_example.py` to include comprehensive steering configurations and improved error handling for guidance steering. - Added functionality to create and demonstrate various steering configurations, including no steering, resampling only, and guidance steering. - Enhanced statistical analysis and visualization of Cα-Cα distances for disulfide bridge pairs across different steering methods. - Introduced new binary images for output visualization of steering comparisons. - Refactored `run_guidance_steering_comparison.py` to support dynamic parameter adjustments for guidance strength and particle counts during experiments.

Co-authored-by: Sarah Lewis <[email protected]>

…into luwinkler/cli_steering

Co-authored-by: Sarah Lewis <[email protected]>

…into luwinkler/cli_steering

…png` and `guidance_visualization_display.png` to clean up the repository.

…longer needed

….py` as it is no longer needed.

…iser/em.yaml` as it is no longer needed.

…nfig/steering/guidance_steering.yaml` as it is no longer needed.

ludwigwinkler · 2025-11-04T15:26:24Z

This is a pull request ready for review for the steering capabilies with BioEmu.

The notebooks directory only contains analysis notebooks and shouldn't really be considered yet for review. There a number of notebooks useful for debugging but not interesting for end users. But I'd like to keep the additional notebooks up until we merge it definitely into the main branch. We also reconfigure them into how-to notebooks.
There is a new src/config/steering directory that showcases the yaml configuration files for the potentials.

We support the following steering functionality:

Only resampling, as guidance/gradient-based steering samples the wrong target distribution.
We can define sampling intervals (start -> end) in which we resample with a certain resampling_freq(uency) between the particles for each sample.
late_steering samples [N] samples up to the start time and only then expands each sample with P particles for a batch size of [N * P].
The potentials are for chain breaks and chain clashes.

The main updates are within the denoiser.py and the steering.py files.
denoiser.py' implements the steering logic at the end of each step of the stochastic DPM solver. steering.py` contains the elementary function we need for steering.

YuuuXie

Thanks for the great efforts! I left some comments, and still need to take a closer look at the steering code, potential definitions etc.

Is there a plan of splitting the current one into two PRs as Sarah suggested? Or this is already one of them?

YuuuXie · 2025-11-07T08:50:24Z

src/bioemu/config/denoiser/dpm.yaml

-N: 50
-noise: 0.0
+N: 100
+noise: 0.5 # original dpm =0 for ode


Wonder if you find when the noise is increased to 0.5 it does not work with 50 steps?

I'd suggest to keep the original setting (N=50, noise=0) as default. On the other hand, we can override the denoiser setting in bioemu.yaml if we use it as a steering example

src/bioemu/denoiser.py

YuuuXie · 2025-11-07T09:03:14Z

src/bioemu/config/steering/chignolin_steering.yaml

+  target: 1.5
+  flatbottom: 0.1
+  slope: 3.0
+  linear_from: 0.5


what is the guiding rule of setting up the parameter linear_from?

YuuuXie · 2025-11-07T09:09:39Z

src/bioemu/convert_chemgraph.py

 from pathlib import Path
+import os

+from matplotlib.pylab import f


remove unused imports

It'll be helpful to run pre-commit checks - those unused imports should be removed by that automatically

YuuuXie · 2025-11-07T09:25:33Z

src/bioemu/convert_chemgraph.py

+    violations = {
+        "ca_ca": ca_seq_distances,
+        "cn_seq": cn_seq_distances,
+        "rest_distances": 10 * rest_distances,


Since you have different units here, it'll be clearer to add unit to the keys, such as ca_ca_nm, cn_seq_nm and rest_distances_angstrom.

Also, do we want to keep this file saving, or this just used for debugging?

YuuuXie · 2025-11-07T14:29:06Z

src/bioemu/steering.py

+
+
+@torch.enable_grad()
+def potential_gradient_minimization(x, potentials, learning_rate=0.1, num_steps=20):


Shall we remove it if we don't have plan to include in this release?

YuuuXie · 2025-11-07T14:31:13Z

src/bioemu/steering.py

+
+import torch.autograd.profiler as profiler
+
+plt.style.use("default")


It's better to move plotting into some analysis scripts instead of in the source code?

YuuuXie · 2025-11-07T14:31:46Z

src/bioemu/steering.py

+    return apply_rotvec_to_rotmat(R, -(sigma_t**2) * score, tol=sde.tol)
+
+
+def stratified_resample_slow(weights):


Do we still want to keep this stratified_resample_slow given we have the "non-slow" version below?

YuuuXie · 2025-11-07T14:43:11Z

src/bioemu/steering.py

+    indices = torch.multinomial(
+        resample_prob, num_samples=num_resamples, replacement=True
+    )  # [BS, num_fk_samples]


Shall we replace this line of multinomial with stratified_resampling?

YuuuXie · 2025-11-07T14:44:53Z

src/bioemu/steering.py

+            torch.ones(BS * num_resamples, device=batch.pos.device) / num_fk_samples
+        )
+    else:
+        resampled_log_weights = None


Shall we remove those log_weights since they are not used? Also if we want to implement some other steering algorithms it might be implemented differently, so maybe we don't make it too specific here?

sarahnlewis

Thanks for the new feature! I haven't finished reviewing yet but am leaving some comments on what I looked at so far.

sarahnlewis · 2025-11-13T12:57:59Z

.gitignore

+*amlt*
+*outputs*
+*cache*
+notebooks/**out**


nit, I suggest you clean up your local 'notebooks' directory rather than putting these specific filenames and wildcards in the .gitignore

sarahnlewis · 2025-11-13T12:59:28Z

README.md

 ## Table of Contents
 - [Installation](#installation)
 - [Sampling structures](#sampling-structures)
+- [Steering for Enhanced Physical Realism](#steering-for-enhanced-physical-realism)


nit, 'steering to avoid chain breaks and clashes' would be more informative

README.md

sarahnlewis · 2025-11-13T16:37:42Z

README.md

+
+- `num_steering_particles`: Number of particles per sample (1 = no steering, >1=steering)
+- `steering_start_time`: When to start steering (0.0-1.0, default: 0.0)
+- `steering_end_time`: When to stop steering (0.0-1.0, default: 1.0)


Are these the values that work best? In other codebases it seems standard to start after 0.0 and stop before 1.0.

sarahnlewis · 2025-11-13T16:38:28Z

README.md

+- `num_steering_particles`: Number of particles per sample (1 = no steering, >1=steering)
+- `steering_start_time`: When to start steering (0.0-1.0, default: 0.0)
+- `steering_end_time`: When to stop steering (0.0-1.0, default: 1.0)
+- `resampling_freq`: How often to resample particles (default: 1)


Is this an interval or a frequency? Does resampling_freq=3 mean resampling after every 3 denoising steps?

If resampling_freq=n means resampling every n denoising steps then this setting is inverse of frequency, and is misnamed. You could call it resampling_interval instead.

sarahnlewis · 2025-11-13T16:46:51Z

notebooks/hydra_run.py

+    torch.cuda.manual_seed_all(SEED)
+
+
+@hydra.main(config_path="../src/bioemu/config", config_name="bioemu.yaml", version_base="1.2")


I suggest removing this file.

sarahnlewis · 2025-11-13T16:47:46Z

notebooks/load_md.py

@@ -0,0 +1,202 @@
+# Working with Blob Storage in Feynman/EMU


This is internal stuff that doesn't belong in this repo.

sarahnlewis · 2025-11-13T16:48:41Z

notebooks/physicality_steering_comparison.py

+                overrides=[
+                    "num_samples=35",
+                    "steering.late_steering=false",
+                    # "sequence=GYDPETGTWG",


Please remove commented code

sarahnlewis · 2025-11-13T16:49:20Z

notebooks/potential_functions.py

+plt.style.use('default')
+
+
+def potential_loss_fn(x, target, tolerance, slope, max_value, order):


What's this file for?

sarahnlewis · 2025-11-13T16:50:45Z

notebooks/run_guidance_steering_comparison.py

+    print(f"Sampling completed. Data kept in memory.")
+
+    # Clean up temporary directory
+    if os.path.exists(temp_output_dir):


Would tempfile.TemporaryDirectory do instead?

microsoft-github-policy-service · 2025-11-14T08:57:00Z

@ludwigwinkler please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.

@microsoft-github-policy-service agree [company="{your company}"]

Options:

(default - no company specified) I have sole ownership of intellectual property rights to my Submissions and I am not making Submissions in the course of work for my employer.
@microsoft-github-policy-service agree
(when company given) I am making Submissions in the course of work for my employer (or my employer has intellectual property rights in my Submissions by contract or applicable law). I have permission from my employer to make Submissions and enter into this Agreement on behalf of my employer. By signing below, the defined term “You” includes me and my employer.
@microsoft-github-policy-service agree company="Microsoft"

Contributor License Agreement

Contribution License Agreement

This Contribution License Agreement (“Agreement”) is agreed to by the party signing below (“You”),
and conveys certain license rights to Microsoft Corporation and its affiliates (“Microsoft”) for Your
contributions to Microsoft open source projects. This Agreement is effective as of the latest signature
date below.

Definitions.
“Code” means the computer software code, whether in human-readable or machine-executable form,
that is delivered by You to Microsoft under this Agreement.
“Project” means any of the projects owned or managed by Microsoft and offered under a license
approved by the Open Source Initiative (www.opensource.org).
“Submit” is the act of uploading, submitting, transmitting, or distributing code or other content to any
Project, including but not limited to communication on electronic mailing lists, source code control
systems, and issue tracking systems that are managed by, or on behalf of, the Project for the purpose of
discussing and improving that Project, but excluding communication that is conspicuously marked or
otherwise designated in writing by You as “Not a Submission.”
“Submission” means the Code and any other copyrightable material Submitted by You, including any
associated comments and documentation.
Your Submission. You must agree to the terms of this Agreement before making a Submission to any
Project. This Agreement covers any and all Submissions that You, now or in the future (except as
described in Section 4 below), Submit to any Project.
Originality of Work. You represent that each of Your Submissions is entirely Your original work.
Should You wish to Submit materials that are not Your original work, You may Submit them separately
to the Project if You (a) retain all copyright and license information that was in the materials as You
received them, (b) in the description accompanying Your Submission, include the phrase “Submission
containing materials of a third party:” followed by the names of the third party and any licenses or other
restrictions of which You are aware, and (c) follow any other instructions in the Project’s written
guidelines concerning Submissions.
Your Employer. References to “employer” in this Agreement include Your employer or anyone else
for whom You are acting in making Your Submission, e.g. as a contractor, vendor, or agent. If Your
Submission is made in the course of Your work for an employer or Your employer has intellectual
property rights in Your Submission by contract or applicable law, You must secure permission from Your
employer to make the Submission before signing this Agreement. In that case, the term “You” in this
Agreement will refer to You and the employer collectively. If You change employers in the future and
desire to Submit additional Submissions for the new employer, then You agree to sign a new Agreement
and secure permission from the new employer before Submitting those Submissions.
Licenses.

Copyright License. You grant Microsoft, and those who receive the Submission directly or
indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license in the
Submission to reproduce, prepare derivative works of, publicly display, publicly perform, and distribute
the Submission and such derivative works, and to sublicense any or all of the foregoing rights to third
parties.
Patent License. You grant Microsoft, and those who receive the Submission directly or
indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license under
Your patent claims that are necessarily infringed by the Submission or the combination of the
Submission with the Project to which it was Submitted to make, have made, use, offer to sell, sell and
import or otherwise dispose of the Submission alone or with the Project.
Other Rights Reserved. Each party reserves all rights not expressly granted in this Agreement.
No additional licenses or rights whatsoever (including, without limitation, any implied licenses) are
granted by implication, exhaustion, estoppel or otherwise.

Representations and Warranties. You represent that You are legally entitled to grant the above
licenses. You represent that each of Your Submissions is entirely Your original work (except as You may
have disclosed under Section 3). You represent that You have secured permission from Your employer to
make the Submission in cases where Your Submission is made in the course of Your work for Your
employer or Your employer has intellectual property rights in Your Submission by contract or applicable
law. If You are signing this Agreement on behalf of Your employer, You represent and warrant that You
have the necessary authority to bind the listed employer to the obligations contained in this Agreement.
You are not expected to provide support for Your Submission, unless You choose to do so. UNLESS
REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING, AND EXCEPT FOR THE WARRANTIES
EXPRESSLY STATED IN SECTIONS 3, 4, AND 6, THE SUBMISSION PROVIDED UNDER THIS AGREEMENT IS
PROVIDED WITHOUT WARRANTY OF ANY KIND, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTY OF
NONINFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
Notice to Microsoft. You agree to notify Microsoft in writing of any facts or circumstances of which
You later become aware that would make Your representations in this Agreement inaccurate in any
respect.
Information about Submissions. You agree that contributions to Projects and information about
contributions may be maintained indefinitely and disclosed publicly, including Your name and other
information that You submit with Your Submission.
Governing Law/Jurisdiction. This Agreement is governed by the laws of the State of Washington, and
the parties consent to exclusive jurisdiction and venue in the federal courts sitting in King County,
Washington, unless no federal subject matter jurisdiction exists, in which case the parties consent to
exclusive jurisdiction and venue in the Superior Court of King County, Washington. The parties waive all
defenses of lack of personal jurisdiction and forum non-conveniens.
Entire Agreement/Assignment. This Agreement is the entire agreement between the parties, and
supersedes any and all prior agreements, understandings or communications, written or oral, between
the parties relating to the subject matter hereof. This Agreement may be assigned by Microsoft.

sarahnlewis · 2025-11-14T09:08:44Z

src/bioemu/convert_chemgraph.py

 import torch
 from scipy.spatial import KDTree

+# No wandb logging needed


Suggested change

# No wandb logging needed

sarahnlewis · 2025-11-14T09:10:22Z

src/bioemu/convert_chemgraph.py

    return atom37_bb_pos, atom37_mask


+def batch_frames_to_atom37(pos, rot, seq):


Please add type hints in this file. Do we need both versions of this function?

sarahnlewis · 2025-11-14T09:11:29Z

src/bioemu/convert_chemgraph.py

+    }
+    path = str(Path(".").absolute()) + "/outputs/analysis"
+    np.savez(path, **violations)
+    # data = np.load(os.getcwd()+'/outputs/analysis.npz'); {key: data[key] for key in data.keys()}


Delete commented code

sarahnlewis · 2025-11-14T09:12:36Z

src/bioemu/convert_chemgraph.py

-        pos=pos_angstrom[0],
-        node_orientations=node_orientations[0],
+        pos=pos_angstrom[-1],
+        node_orientations=node_orientations[-1],


… with new dependencies, ensure output directory creation in convert_chemgraph.py, and modify import statements in sample.py for improved script execution.

… and evaluation

…d deprecated steering parameters from YAML files, streamlined steering logic in the codebase, and enhanced documentation to reflect changes in steering functionality and usage. Updated .gitignore to exclude additional files and improved performance of batch processing functions.

…py and clean up logging output for cached embeddings.

…m and node_orientations instead of the last, ensuring correct data is written to .pdb files.

…ewers

… for start and end times.

sarahnlewis · 2026-01-08T11:34:51Z

.gitignore

+.cursor/
+.cursor/** */
+docs/*
+uv.lock


sarahnlewis · 2026-01-08T11:36:30Z

.gitignore

+uv.lock
+
+# samples
+*.pdb


This is problematic because there are .pdb files in the repo, referenced in tests. Instead of adding all this to .gitignore I suggest you don't save samples to the repo directory in the first place. Same comment for many of the other things added here.

sarahnlewis · 2026-01-08T11:39:43Z

README.md

+- `num_steering_particles`: Number of particles per sample (1 = no steering, >1=steering)
+- `steering_start_time`: When to start steering (0.0-1.0, default: 0.0)
+- `steering_end_time`: When to stop steering (0.0-1.0, default: 1.0)
+- `resampling_freq`: How often to resample particles (default: 1)


If resampling_freq=n means resampling every n denoising steps then this setting is inverse of frequency, and is misnamed. You could call it resampling_interval instead.

sarahnlewis · 2026-01-08T11:41:06Z

README.md

+
+### Key steering parameters
+
+- `num_steering_particles`: Number of particles per sample (1 = no steering, >1 enables steering)


The docs should call out that num_steering_particles=n means you will use slightly more n times as much compute to get the same number of samples.

sarahnlewis · 2026-01-08T11:42:02Z

pyproject.toml

    "typer",
    "uv",
+    "einops",
+    "matplotlib>=3.10.7",


Do you need matplotlib and pandas or were these for notebooks that you removed from the PR?

sarahnlewis · 2026-01-08T12:13:08Z

src/bioemu/steering.py

+    score: torch.Tensor,
+) -> torch.Tensor:
+    """
+    Compute x_0 given x_t and score.


nit

Suggested change

Compute x_0 given x_t and score.

Compute expected value of x_0 using x_t and score.

sarahnlewis · 2026-01-08T12:14:22Z

src/bioemu/steering.py

+    """
+    Compute R_0 given R_t and score.
+    """
+    alpha_t, sigma_t = sde.mean_coeff_and_std(x=R, t=t, batch_idx=batch_idx)


hm, is this legit when R lives in SO(3)? What do we need it for anyway? I thought the potentials were all defined in terms of CA positions.

sarahnlewis · 2026-01-08T12:18:50Z

src/bioemu/steering.py

+    return x0_t, R0_t
+
+
+def log_physicality(pos, rot, sequence):


Please add type hints and for consistency with the codebase, use logger, not print.

Function below also needs type hints.

sarahnlewis · 2026-01-08T13:19:34Z

src/bioemu/denoiser.py

+from bioemu.so3_sde import SO3SDE, apply_rotvec_to_rotmat
+from bioemu.steering import get_pos0_rot0, resample_batch

 TwoBatches = tuple[Batch, Batch]


If not using TwoBatches any more, remove its definition

sarahnlewis · 2026-01-08T13:30:39Z

src/bioemu/denoiser.py

+                        "Final Resampling [BS, FK_particles] back to BS, with real x0 instead of pred x0."
+                    )
+                    seq_length = len(batch.sequence[0])
+                    x0 = batch.pos.view(batch.batch_size, seq_length, 3).detach()


This overwrites your list x0 from line 27, no?

ludwigwinkler and others added 28 commits July 21, 2025 14:16

feat: add TODO for fk steering implementation in heun_denoiser

3b3dd46

Merge branch 'main' into luwinkler/fk_steering

3069e93

gitignore test outputs and test case for steering

6829f31

Merge branch 'luwinkler/fk_steering' of os.github.com:microsoft/bioem…

ec98369

…u into luwinkler/fk_steering

first prototype

f61a0aa

feat: implement Kabsch alignment and enhance steering functionality w…

80a1de2

…ith FK sampling

feat: add wandb to .gitignore to exclude Weights and Biases files

046aeaf

Update .gitignore to exclude fasta files in notebooks directory

135b6ef

Update .gitignore to exclude output files from notebooks directory

aea00cd

Remove .cursor/ from tracking and add to .gitignore

0d3dd81

Delete .cursor/rules directory

206860e

Update .gitignore to exclude all files in .cursor directory

f0fee56

Merge branch 'luwinkler/fk_steering' of github.com:microsoft/bioemu i…

fc5093d

…nto luwinkler/fk_steering

first workin prototype of fast_steering

5452976

Update .gitignore to exclude documentation files

516046d

- Added an entry to .gitignore to ignore all files in the docs directory, preventing them from being tracked by Git.

ludwigwinkler self-assigned this Oct 15, 2025

ludwigwinkler changed the title ~~Luwinkler/cli steering~~ luwinkler: Steering for BioEmu Oct 15, 2025

ludwigwinkler and others added 11 commits November 4, 2025 16:00

Update README.md

9e6c71b

Co-authored-by: Sarah Lewis <[email protected]>

refs for readme on steering and smc

65f44a0

Merge branch 'luwinkler/cli_steering' of github.com:microsoft/bioemu …

4c09554

…into luwinkler/cli_steering

Update notebooks/README_hydra_run.md

b379b2e

Co-authored-by: Sarah Lewis <[email protected]>

some renaming files and typos

ac46972

Merge branch 'luwinkler/cli_steering' of github.com:microsoft/bioemu …

da161c3

…into luwinkler/cli_steering

Remove unused guidance images: deleted `guidance_steering_comparison.…

7aaa635

…png` and `guidance_visualization_display.png` to clean up the repository.

Remove profiler module: deleted src/bioemu/profiler.py as it is no …

9a2a405

…longer needed

Remove unused steering scratch pad: deleted `src/steering_scratch_pad…

25b5bfd

….py` as it is no longer needed.

Remove unused denoiser configuration: deleted `src/bioemu/config/deno…

2c97e05

…iser/em.yaml` as it is no longer needed.

Remove unused guidance steering configuration: deleted `src/bioemu/co…

da475bf

…nfig/steering/guidance_steering.yaml` as it is no longer needed.

Merge branch 'main' into luwinkler/cli_steering

16c2da8

ludwigwinkler marked this pull request as ready for review November 4, 2025 15:31

YuuuXie reviewed Nov 7, 2025

View reviewed changes

sarahnlewis reviewed Nov 13, 2025

View reviewed changes

sarahnlewis changed the title ~~luwinkler: Steering for BioEmu~~ feat: steering to avoid chain breaks and clashes Nov 14, 2025

sarahnlewis reviewed Nov 14, 2025

View reviewed changes

ludwigwinkler added 10 commits December 3, 2025 13:14

Update .gitignore to exclude additional files, enhance pyproject.toml…

741d69d

… with new dependencies, ensure output directory creation in convert_chemgraph.py, and modify import statements in sample.py for improved script execution.

simplified steering logic and first prototype for integrated sampling…

e28c475

… and evaluation

cleaned up notebooks and configs

4d6455e

Remove commented TODO regarding embedding file copying in get_embeds.…

46161fd

…py and clean up logging output for cached embeddings.

fix formatting

480b405

Fix save_pdb_and_xtc function to use the first element of pos_angstro…

9252860

…m and node_orientations instead of the last, ensuring correct data is written to .pdb files.

fixing small deviations in the code to reduce mental load of the revi…

885b4f2

…ewers

fixing diverging code

479d84b

Update steering parameters in README.md to reflect new default values…

0ac2b42

… for start and end times.

sarahnlewis reviewed Jan 8, 2026

View reviewed changes



		@torch.enable_grad()
		def potential_gradient_minimization(x, potentials, learning_rate=0.1, num_steps=20):


		import torch.autograd.profiler as profiler

		plt.style.use("default")

		return apply_rotvec_to_rotmat(R, -(sigma_t*2) score, tol=sde.tol)


		def stratified_resample_slow(weights):

		torch.cuda.manual_seed_all(SEED)


		@hydra.main(config_path="../src/bioemu/config", config_name="bioemu.yaml", version_base="1.2")

		plt.style.use('default')


		def potential_loss_fn(x, target, tolerance, slope, max_value, order):

		return atom37_bb_pos, atom37_mask


		def batch_frames_to_atom37(pos, rot, seq):


		### Key steering parameters

		- `num_steering_particles`: Number of particles per sample (1 = no steering, >1 enables steering)

	Compute x_0 given x_t and score.
	Compute expected value of x_0 using x_t and score.

feat: steering to avoid chain breaks and clashes #171

Are you sure you want to change the base?

feat: steering to avoid chain breaks and clashes #171

Uh oh!

Conversation

ludwigwinkler commented Oct 15, 2025

Uh oh!

ludwigwinkler commented Nov 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

YuuuXie left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sarahnlewis left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

microsoft-github-policy-service bot commented Nov 14, 2025

Contribution License Agreement

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

ludwigwinkler commented Nov 4, 2025 •

edited

Loading