Feat: Add NBFix and MBAR estimator #66

WangXinyan940 · 2022-10-26T07:33:00Z

Refactor:

Split old api.py to separate files. Now the api.py only contains Hamiltonian class. The generators are moved to dmff/generators/*.py

New Features:

LennardJonesForce and NBFix support.
MBAR estimator for free energy estimation and reweighting.

Bug fix:

Use template name instead of residue name for atomic charge matching.

* add nblist wrapper and its docs * fix(generator): existing jax bond/angle generators * fix: LJ bug; all classical code can be jitted; bug at test_gaff2_force * fix test_gaff2_force bug * fix: non-differentiable error, move args check in the api.py * fix: remove redundancy `box=jnp.array(box)`; confirm isinstance_jnp is jit-compatiable * refine(pme): code prettify * refine(utils): code prettify * add(nblist): add jit and auto update nblist * refine(classical): rm unused import * refactor(classical): add dispCorr * add(classical): add free energy support * refactor renderXML with clean commits (#25) * feat:rewrite renderXML with very clean commit history * fix bug in Torsion renderXML * add test_utils as a placehold * fix: fix typo in api.py * docs: add renderXML related api usages * refine(fep): rm debugging codes * add(cicd): CI/CD workflows * add(ut): ut for regularize_pairs and buffer_scales * fix(CI/CD): activate conda env * fix(CI/CD): wrong number in test_nblist * add(requirement): add dependencies list * add(CI/CD): unittest workflows * add(requirements): dependencies list * add test_utils.py * fix: modified unit test results * add `r` to avoid latex being recognized as an escape character * fix(ut): wrong number in test_nblist * Fix unit test related problems (#27) * add test_utils.py * fix: modified unit test results * add `r` to avoid latex being recognized as an escape character * fix(ut): wrong number in test_nblist Co-authored-by: Yingze Wang <ericwangyz@pku.edu.cn> * refine(ut): code prettify in test_nblist * Add CI/CD Workflows (#26) * add(CI/CD): unittest workflows * add(requirements): dependencies list * refine(ut): code prettify in test_nblist * update(api): switch default dispcorr to False * Chore: clean admp module up * chore: clean up classical and api.py * fix(ut): withdraw last commit about fix wrong number in test_nblist * update(api): fix dispcorr countmat bugs * update(pme): code prettify & gmx ewald coeff determine * add(constants): module to control constants * add(unittest): fep ut * update(test_nblist): unused imports * update(gitignore): acpype cache * refactor(test_classical): split to several files * refactor(test_classical): rename test_classical * fix(inter): PME in classical forcefield * update(fep): use dmff.common.constants * update(pme): default args in setup_ewald_parameters * Fix the nb list description in doc * update mkdocs.yml and a simple test of api generator * fix: add step_pol arg in ADMPPmeForce __init__ to fix it can not be jitted bug * add step_pol arg in ADMPPmeForce __init__ to fix it can not be jitted bug * Improve docstrings in sgnn * add read_input_info in dmff.admp.parse.py to deal with info in pdb and xml (input:pdb,xml output: multipoles/polarizabilities/tholes...) use forcefield.xml instead of other xmls, add read_admp_xml in parser in order to read the new form of xml file change line 795 in pme.py and add distribute_matrix in pairwise.py in order to wrap a jit outside my admp_calculator * Update some examples and ref_outs * feat: auto gen docs refs * fix: typo in requirements.txt * add change leading terms in api.py * add fluctuated leading term computation in dmff.api * add fluctuated leading term compute in dmff.api * use dmff.api to deal with input ; wrap admp_calculator to change fixed params into fluctuated one amf jit * wrap compute leading term in run.py * add a markdown and a generate_calculator outside admp_calculator function * feat(classical): Hamiltonian can create total energy function now. (#33) * Docs refine (#34) * add(CI/CD): unittest workflows * add(requirements): dependencies list * update(doc): fix bad render of \hat * update(doc): add examples and doc struct refine * Improve the fluctuating charge jupyter notebook demo * Delete parser_bk.py * Fix the fluctuating atom charge demo * feat: update docs configs * fix: fix test_nblist bug * update: license in docs * fix: import missing in api.py (may caused by formatter) * FIX: wrong links in docs and add requirements (#37) * update(doc): fix bad render of \hat * update(doc): add examples and doc struct refine * hotfix(docs): license bugs Co-authored-by: Kuang Yu <windwhisper.yu@gmail.com> Co-authored-by: Roy Kid <lijichen365@126.com> Co-authored-by: Jichen Li <42854324+Roy-Kid@users.noreply.github.com> Co-authored-by: crone <2223469329@qq.com> Co-authored-by: WangXinyan940 <44082181+WangXinyan940@users.noreply.github.com>

Merge devel into master

KuangYu

See review

dmff/mbar.py

KuangYu · 2022-10-31T12:32:12Z

dmff/mbar.py

+            return eff_samples, state_effect
+        return eff_samples
+
+    def estimate_free_energy(self, unew):


We should consider computing free energies between two target states.

Target states and params should be passed in as argument.

…ld jump between 12 and 13)

Ericwang6 · 2022-11-05T06:11:39Z

I've removed some unused imports

…f using single frame

KuangYu

See comments

KuangYu · 2022-11-09T13:01:10Z

examples/mbar/demo.ipynb

+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "class OMMNPTSampleState(SampleState):\n",


Should we include this in mbar.py as part of standard package?
Also, When the name indicates it's NPT, the pressure seems to be hard coded.

KuangYu · 2022-11-09T13:05:32Z

dmff/optimize.py

+                                        decay_rate=0.99)
+
+    # Combining gradient transforms using `optax.chain`.
+    chain = [optax.adam(scheduler), optax.clip(clip)]


Seems that the optimizer is hard coded in here, while I am not sure if ADAM is the optimal in here, maybe we should consider allowing the users to specify optimizers. Maybe add a tag in argument, like: optimizer="ADAM"

Ericwang6 and others added 12 commits June 14, 2022 17:58

Merge pull request #39 from deepmodeling/devel

9799844

Merge devel into master

Merge pull request #45 from deepmodeling/devel

af4b1f2

Merge devel into master

Merge remote-tracking branch 'origin/feat-covmap' into feat-nbfix

0e109db

Merge remote-tracking branch 'origin/feat-covmap' into feat-nbfix

d1fe0e5

Merge remote-tracking branch 'origin/feat-covmap' into feat-nbfix

f17ae2b

Refactor: decompose api.py and generators to separate files

9348487

Merge branch 'devel' into feat-nbfix

1ef38e2

feat(MBAR): Add differentiable MBAR impl

66562ab

bug fix: record the name of matched template

1b6473e

Add unit test for MBAR estimator

13aa8cb

feat: Estimate free energy of an extra state

41bc501

WangXinyan940 requested review from Ericwang6, KuangYu and Roy-Kid October 26, 2022 07:33

KuangYu requested changes Oct 31, 2022

View reviewed changes

WangXinyan940 and others added 10 commits November 2, 2022 21:01

set specific version no. for jax in installation guide

c30ca0d

Update MBAR Estimator API

e597d6b

Update unit test for the latest MBAR API

97583c5

Update github workflow to support new unit test.

e7f4ed7

Update requirement of pymbar

bcd1390

fix package including problem

15dbdbd

Change cell vector to prevent numerical problem (ceil(12.0 / 1.0) wou…

20f0d3f

…ld jump between 12 and 13)

Update settings.py

dd50e16

remove unused imports

7609886

fix "==" in requirements

a1f5fbe

Ericwang6 and others added 3 commits November 6, 2022 01:30

update (gitignore): hmtff cache

55452aa

update (api): docstring in createPotential

75ecc84

Update optax transforms for force field parameters

fa5ca20

WangXinyan940 added 16 commits November 7, 2022 12:08

Update __init__.py

d5a1aea

Merge branch 'feat-nbfix' into feat-optimizer

5d77215

Update LJ jax force API in LennardJonesForce generator

1bf8710

Merge branch 'feat-nbfix' into feat-optimizer

932d4ef

Let energy function in TargetState use the whole trajectory instead o…

b75682e

…f using single frame

Update MBAR UT to fit API change

585ba81

Merge branch 'feat-nbfix' into feat-optimizer

adcfcc2

Remove the using of numpy to make free energy & weight differentiable

fab194f

Merge branch 'feat-nbfix' into feat-optimizer

8358fc7

Update __init__.py

d1e6d79

Set precision again in dmff.mbar after importing pymbar

6ca6664

Merge branch 'feat-nbfix' into feat-optimizer

9e0bc28

Increase MBAR numerical stability

e4e6b7a

Merge branch 'feat-nbfix' into feat-optimizer

6eedcfe

Update an example of MBAR-based optimization

b2eb539

Correct the typo in document

5c45204

KuangYu requested changes Nov 14, 2022

View reviewed changes

WangXinyan940 added 3 commits November 14, 2022 11:28

Add openmm sample state.

e0cc8fc

Update genOptimizer to support multiple optimizers

2e7adaa

Set default pressure to be 0 (NVT)

f3e220b

KuangYu approved these changes Nov 14, 2022

View reviewed changes

WangXinyan940 merged commit 8395f68 into devel Nov 14, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Feat: Add NBFix and MBAR estimator #66

Feat: Add NBFix and MBAR estimator #66

Uh oh!

WangXinyan940 commented Oct 26, 2022

Uh oh!

KuangYu left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

KuangYu Oct 31, 2022

Uh oh!

Ericwang6 commented Nov 5, 2022

Uh oh!

KuangYu left a comment

Uh oh!

KuangYu Nov 9, 2022

Uh oh!

KuangYu Nov 9, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Feat: Add NBFix and MBAR estimator #66

Feat: Add NBFix and MBAR estimator #66

Uh oh!

Conversation

WangXinyan940 commented Oct 26, 2022

Uh oh!

KuangYu left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

KuangYu Oct 31, 2022

Choose a reason for hiding this comment

Uh oh!

Ericwang6 commented Nov 5, 2022

Uh oh!

KuangYu left a comment

Choose a reason for hiding this comment

Uh oh!

KuangYu Nov 9, 2022

Choose a reason for hiding this comment

Uh oh!

KuangYu Nov 9, 2022

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants