NIAID Guidehouse/MSC contract May 1, 2022 - Feb 28, 2023

1. General ChimeraX improvements to support NIAID.
   1. Specific improvements to assist NIAID personnel to transition away from the unsupported legacy Chimera program, e.g.:
      1. Join disconnected peptide segments via a new peptide bond
      2. Renumber residues
      3. Rename chain identifiers
      4. Provide both PDB and UniProt residue numberings
      5. Prepare structures for input to docking programs (remove alternate atom locations, add missing side chains, etc.)
      6. Energy minimize atomic structures
      7. More GUIs (specifically copy/combine)

2. Support the NIH 3D pipeline development, including any changes to ChimeraX to support ongoing development.
   1. Revamp NIH 3D scripts to use new presets
   2. As the pipeline code matures and reaches deployment, support for any changes to NIH 3D scripts, or ChimeraX itself, needed for optimal pipeline utility.
      1. Improved GLTF output to automatically center models at origin.
      2. Tested new PDB assembly single model mmCIF files.
   3. Perform early testing on the NIH 3D web site and provide feedback.

3. Virtual reality for molecular structure, cryoEM and medical imaging.
   1. Setup mixed reality video capture at BioVizLab on NIH campus.
   2. Allow meeting participants to add models from local files during a VR session.
      1. For this round, this may be a minimal viable project – perhaps for small files only.
   3. Add a VR toolbar in the ChimeraX interface.
   4. Add a method to save sessions in VR.
   5. Improve visualization of 3D medical imaging data in VR.
      1. Improved interface for manipulating the visualization – e.g. Medical Imaging Toolbar
   6. Support making or editing interactive 3d image segmentations in VR.
   7. Provide local interactive molecular dynamics in VR.
   8. Create video tutorials for how to use multi-person VR.
   9. Hold VR meeting office hours to train researchers to use VR capabilities.
   10. Other VR improvements.

4. CryoEM modeling utilizing machine learning structure prediction.
   1. Provide a structure prediction service (e.g. AlphaFold or RoseTTAFold) that can handle more than the 1000 amino acids possible on Google Colab.
   2. Provide visualization of machine learning structure quality metrics in 2D plots and on the structure itself to evaluate correctness of domain packing.
   3. Provide ability to specify structure templates when making predictions.
   4. Provide more reliable multiple sequence alignments (e.g. with mmseqs2) to allow structure predictions where current sequence alignment (e.g. with hhblits) fails.
   5. Provide reliable energy minimization of predicted structures.
   6. Other cryoEM machine learning structure prediction.

5. DICOM
   1. Improve DICOM reader by including more data types such as segmentations, lossless JPEG used in clinical scans, and 16 bit image data.
   2. Investigate implementation of a DICOM metadata browser

6. Outreach
   1. Instructional material and tools documentation.
      1. Detailed instructions for all features shall be provided in a user manual.
      2. Written user guides and tutorials shall be available as an HTML page.
      3. Create videos demonstrating new capabilities.
      4. Present tutorials at webinars and workshops training to use new capabilities
   2. Attendance at meetings or workshops as required by NIAID

7. Administration
   1. Submit monthly written reports of accomplishments

Meeting Minutes

6/9/2022

Phil, Philip, Michal Stolarczyk and David Liou from 3D pipeline team, Meghan (last half); Zach, Eric, Elaine, Greg, TomG, Scooter

Self-introductions from both teams.

Michal is currently the primary developer working on NIH3D, working for last month on infrastructure. ChimeraX script has been running without issues, so he hasn't been working on that part. ~10-min slide presentation.

Running Ubuntu 20.4 0 within EC2 instance within VPC within AWS. Involves Prefect and GraphQL.

Recent updates:

• workflow deployment: standardized flow names, external Postgres database
• workflow handling: ​https://niaid.github.io/nih_3D_workflows, JSON-based configuration

Planned enhancements:

• switch to ECS for running flows (to handle more at the same time)
• increase reproducibility of results by including code versioning
• use newer version of NIH3D presets provided by Eric

Phil: Eric please explain unified script and presets

Eric: originally all the types of view/displays that could be requested by the user were in a giant script with lots of commands, and we decided to refactor that as a set of presets where each preset generates one type of view. Now a smaller script calls the individual presets as needed. This set of presets is available as a bundle on the ChimeraX toolshed. Would have to re-install if ChimeraX is not persistent between runs.

Michal: ChimeraX is persistent on the same machine, but different machines are used. When we use ECS it will deploy a docker image that could have the bundle already installed.

Eric: how long to deal with my pull request? Michal: about a couple of weeks.

Eric: ChimeraX now has "renumber" command, on the way towards having a GUI to do it.

Some discussion of reproducibility, ChimeraX preferences and fetch caching.

Phil: are ChimeraX-calculated values saved somehow? Eric: currently, the script sends them to standard-out with a standardized prefix to facilitate grepping the lines. Could be changed if needed.

TomG: new "alphafold contacts" command to draw pseudobonds as explained in https://www.rbvi.ucsf.edu/chimerax/data/alphafold-contacts-jun2022/afcontacts.html

Phil: what about similar pseudobonds for contacts? Elaine: can be done with "contacts" command but we don't have anything to color them continuously by length yet. Scooter: might be interesting to show histogram by length with "crosslinks histogram". Elaine: clicking a histogram bar interacts with corresponding pseudobonds.

Zach: has reached out to Jason Rubinstein, he will take a look at his data using our latest DICOM/VR and get back to us

Meghan: we still need to give feedback on where the buttons are on different VR hand-controllers, for better mapping to the cone display in the headset. TomG: we currently use two different layouts, one for Vive and one for Oculus Rift. So more could be added.

5/26/2022

Phil, Philip, Darrell (Meghan sick today); Zach, Eric, Elaine, Greg, TomG, Scooter

Presentation by new NIH3D workflow developer, Michal? Not scheduled yet, maybe next time. Phil: will see if he's available to join this meeting. Phil: have done a little testing on the latest bundle from the toolshed, will send (minor) comments via e-mail. Main issue this week was NIH3D pipeline breaking due to changes in metadata, but has been dealt with. Darrell: was not related to Chimera, was caused by a change in the XML from EMDB.

Darrell: how can we get an approximate contour level if we fail to get one from EMDB? Eric: there is already some iterative process to setting a contour level that makes a coherent blob. Darrell: sounds good.

Darrell: we would like TomG to come to Wash DC and also Rocky Mtn Labs. Scooter: we budgeted for those two trips. Phil: maybe for the Biovis lab opening events. Darrell: time frame is uncertain, originally planned for fall but there are many other important meetings at about the same time, so it may be delayed. Darrell: will be going to Nigeria for an NIH meeting (the last "H3" meeting; had demoed VR there in 2018). VR demo has been requested for a meeting in Mali, at rather short notice. Phil: there was a VR-collab demo between Mali and Uganda not long ago.

Darrell: what systems are good to show in VR? Phil: TB drug resistance. Is this a demo? Darrell: maybe a video but not by this Sunday when they wanted.

Phil: I've been trying the release candidate from 5/20, haven't found problems.

Darrell: do you have an M1 version? Zach: In progress. Elaine: I've been using the existing download (via emulation) on an M1 machine without problems, works well. Phil: everything seems twice as fast, even with the emulation, on my M1 machine.

TomG: CPU speed on M1 is great. I made a benchmark page https://www.rbvi.ucsf.edu/chimerax/data/czi-nov2021/apple_m1.html

Philip: what are the Rocky Mountain cluster. Darrell: GPUs are the one before A100, something starting with a different letter. Scooter: We are using A40 and A100 on our cluster.

Scooter: would like to discuss DICOM at a later meeting. Elaine: are we hearing from Michal next time? Phil: I'll check with him and get back to you.

Darrell: Brad and Ziv may be able to provide datasets. Scooter: maybe we can also get them to join the discussion. Elaine: to say how they analyze the data. Darrell: focus on visualization component, not to replace these other packages like Slicer3D. Phil: maybe some analyses would be great to combine with VR. Darrell: maybe should get Dave (he's responsible for the Visible Human stuff) to present as well. How about we get them to view DICOM in ChimeraX and give comments? Scooter: that would be good, too. TomG: we need a VR-specific perspective, though. Scooter: maybe NIH folks can set up a DICOM-VR session. Phil: maybe Dave + TomG in VR meeting. TomG: sounds OK, but I still think that we need to get more feedback from Jason Rubinstein (or other people that really work with both medical imaging and VR). Remote meetings have limitations, however; talking face-to-face in the same room is really the best if possible, say during my visits. Darrell: then maybe Jeff Solomon, perhaps NIAID's most DICOM-oriented scientist. Dave certainly knows DICOM, too. Maybe we could fly Jason Rubinstein in from Wisconsin. TomG: can we afford to fly Zach in too? Scooter: currently contract has 2 trips for one person. Darrell: maybe we can find additional funding.

5/12/2022

Phil, Philip, Meghan; Elaine, Eric, Zach, Greg, Scooter

Scooter: we are planning a group off-site get-together, maybe October, maybe sailing.

Scooter: contract agreement is done! Tom G is the unofficial lead on this NIAID funding, but he's on vacation now, so we don't have much to say yet. Zach will be working on the DICOM stuff. Status of NIH 3D?

Phil: last I recall, still waiting on progress with the workflow server to handle new submission process and use of presets. Eric & Greg worked out the toolshed issue, and the bundle is available there. Philip got called to another project and another developer (Mihal?) is taking over the NIH 3D workflow programming. Kristin is now the development coordinator for NIH3D.

Scooter: maybe that person can present at our next meeting? Meghan & Phil agree. Meghan will ask Mihal about doing this.

Scooter: progress on AlphaFold but we would also like to look into RosettaFold calculations too. Phil: RosettaFold is harder to use. It was also reasonable to start with AlphaFold since EBI is providing a database of already-calculated models. Scooter: there was some talk of RCSB starting a RosettaFold database.

Scooter: Elaine's been presenting DICOM and DICOM viewers, and we have a couple of possible people to work with for real data & applications, Jason Rubenstein at Medical College of Wisconsin, and previously there was somebody else in South America. Of course we still want to hear any requests or suggestions your group has for DICOM in VR, which would be top priority for this contract.

Meghan: we've had a lot going on, not to mention the pandemic disruptions, so we need to revisit VR medical applications along with our collaborators in Uganda. We're meeting with them in a couple of weeks. We're also working with other groups in VR medical applications (NHS in UK for health training) and come up with our next strategy and get back to you later. I'm glad this is back on the radar! We will also be getting more hardware.

Scooter: we have a good wifi setup for untethered Quest headset now. Required wrangling with UC IT to allow setting up our own wifi6 router (?) only for this limited purpose. Campus wireless was unusable for this purpose.

Phil: in ChimeraX, a big advantage is that we can go inside models in VR. Medical Holodeck doesn't really allow this, framerate goes way down. Meghan: Medical Holodeck is just too heavyweight for convenient use in many situations.

Phil: we like that the mapping of buttons to the cones shown in VR. However, the mapping is not good for some of the controllers where the buttons are in different places. So we would like Tom G to take a look at making different mappings that better match the different controllers. Elaine: I'll make a ticket and maybe Phil can add comments about which ones are mismatched.

Phil, Meghan: Another recent problem with old (Chimera) pipeline may be related to EMDB metadata. Eric: new (ChimeraX) pipeline gets recommended contour level; don't know what the old pipeline does. Phil: I think the problem is not in Chimera but in some other step.

Scooter: 1.4 release is coming soon, release candidate is available.

4/14/2022

Phil, Philip; Elaine, Eric, TomG, Zach, Greg

Today Darrell authorized "MSC" (or logical equivalent) to act on UCSF’s final proposal against the SOW!

Phil: I hear that the new NIH3D workflow will be deployed soon. Kristin will be project coordinator, maybe she will attend some of our meetings in the future.

Phil: Eric, please add the GLTF (glb) saving option "center 0,0,0" to the workflow. TomG: don't bother, I'll make that the default.

Eric: I still can't upload the bundle to the toolshed, because... Greg: dependency field issue with comma in string. Will be solved by upgrading to newer mariadb, should be done within a week.

Eric is working on joining models, starting with peptide bond.

Phil: gave a talk to the structural biology group, showed alphafold PAE plots -- people were very interested and liked the ability to switch between domain and confidence (pLDDT) colorings. Elaine: there's a Youtube video of TomG's recent SBGrid workshop on that topic.

TomG: maybe we can also color by conservation calculated from the giant sequence alignment alphafold uses to covariance.

Phil: people had several suggestions: (1) to predict every chain in an existing PDB. TomG: you can do that now, just open that structure first and use "alphafold match" ... however, what you get may have a different albeit similar sequence (top BLAT hit) (2) another request was to draw chain boundaries on the PAE plot for the multichain case. (3) for selected residue, color others by their PAE with the selected residue. Could also color-code pseudobonds, or only draw pseudobonds for certain range of PAE values. (4) report X,Y cursor location when over the plot (e.g. residue numbers, chain IDs) TomG: what about on the structure? select the 2 residues? draw a pseudobond between them? (5) allow zooming in to regions within the plot

Phil: does predicting in ChimeraX work with a user's Colab Pro or Pro+ plan? TomG: yes, since you sign in to Google it knows you have an account.

TomG: do you have any experience with Vive focus 3? Phil: we may be getting one soon, as well as HP Reverb and Pico (sp?), which has eye tracking... the hot new headsets. Will let you know. TomG: I like the Vive pro 2, higher resolution, eye tracking.

3/31/2022

Phil, Philip; Elaine, Eric, TomG, Zach, Greg

Status of SOW? Still being negotiated. Phil: what is highest priority? TomG: We're told it won't start until May 1, so we haven't thought mcuh about it yet. The SOW includes machine learning & AlphaFold stuff. Phil: everybody's interested in using that now. TomG: shows AlphaFold "predicted aligned error" (PAE) plot just added to ChimeraX. Google Colab version doesn't use templates, so PAE domain identification is based on residue covariance. PAE plot and interaction with structure in ChimeraX is similar to that available at the AlphaFold Database, except the ChimeraX one is better if you highlight an off-diagonal box on the plot (shows 2 colors on structure).

Discussion of importance/predominance of disordered regions in proteins. Phil: simulations don't model these well.

Phil used to work at Tripos. Elaine: my first modeling work was with SYBYL (from Tripos) on the Evans & Sutherland! Phil: booting up the E & S took forever.

TomG: I added a tug command to fix up linkers after building a large, complicated protein structure.

Elaine: figured out how to show time series in 3D Slicer. At least some of the problem was due to the version of 3D Slicer - time series example files for one version don't work in the other (without additional steps needed to define a set of maps as a time series).

TomG: looked at the NIAID TB-VR experience website, sent feedback to Phil. Phil: here's the URL, public now: ​https://tbportals.niaid.nih.gov/tb-in-vr TomG: "save" GLTF has a "center" option. Maybe we should make the default "center 0,0,0" ... downside is when the user wants to preserve the relative locations of multiple structures.