eval_cases/molecular_vis/eval_analysis_workflows.yaml

# 1. RAS-RAF on a plasma membrane
# The RAS protein is lipid anchored and sits on the inner leaflet of the plasma membrane where it can bind RAF.
# This is a snapshot of a RAS protein bound to the RBD and CRD domains of RAF on top of an eight-component plasma membrane setup with the coarse-grained Martini 2 force field.
# This simulation frame is an example from the following study https://doi.org/10.1073/pnas.2113297119 and can be found online at: https://bbs.llnl.gov/data/ras-lipid-dependent-dynamics-data
- vars:
    question: |
        1. Please load the Martini coarse-grained simulation file from "ras-raf-membrane/data/ras-raf-membrane.gro" into VMD. The simulations has a membrane and a RAS-RAF protein complex.
        
        2. Use VMD to show a zoomed in side view of the membrane and center on the protein with the protein below the membrane.
        For the bilayer only show the PO4 lipids beads and ROH cholesterol bead and color them gray. 
        Also show the protein back bone beads coloring RAS (resid 2 to 187) red and RAF (resid 188 to 329) blue.
        Take a screenshot of the visualization.

        3. Analyze the visualization and answer the following questions:
        Q1: Are there any cholesterol head groups in the bilayer center? (yes/no)

        Q2: How many lipids are there within 1.5 nm of the RAF protein?
          A. 0
          B. 0-3
          C. 3-5
          D. >5

        4. Save your work:
        Save the VMD state as "ras-raf-membrane/results/{agent_mode}/ras-raf-membrane.vmd".
        Save the screenshot of the visualization as "ras-raf-membrane/results/{agent_mode}/ras-raf-membrane.png".
        Save the answers to the analysis questions in plain text as "ras-raf-membrane/results/{agent_mode}/answers.txt".
  assert:
    - type: llm-rubric
      subtype: vision
      value: |
          1. Camera Position: Is the image centered on a protein complex composed of clearly distinguishable half red RAS and half blue RAF components?
          2. Membrane Placement: Is a lipid bilayer clearly visible above the protein?
          3. Correct Leaflet Ordering: Is the inner leaflet of the bilayer closer to the protein and the outer leaflet positioned farther away?
    - type: llm-rubric
      subtype: text
      value: |
          1. Q1 correct answer: Yes
          2. Q2 correct answer: C. 3-5

# Q1 reason: quite some gray dots from cholesterol ROH headgroup bead can be seen in the center of the bilayer, so the answer is YES.
# Q2 reason: if you run the flowing VMD selection command (name PO4 ROH) and within 15 of (index 3457 to 35009 and resid 188 to 329)
# You will find 4 lipids head group beads, so the answer is C. 3-5 
# Note here it’s important to only count each lipid once, e.g. either use a single reference bead per lipid or center of mass, also selecting the protein can be tricky as the lipids and water also if resid’s in the 188 to 329 range so either exclude other molecule or sub select only the protein residues

# 2. Curved membrane simulation
# As part of a protocol paper for building membrane using the coarse-grained Martini 3 force field, https://doi.org/10.1016/bs.mie.2024.03.010, a tutorial was developed: https://bbs.llnl.gov/data/building-membranes-data
# Tutorial 4 is on curved membranes which is used here as an example of bilayer visualization. 
- vars:
    question: |
        1. Please load the Martini coarse-grained simulation file from "curved-membrane/data/curved-membrane.gro" into VMD.

        2. Use VMD to show a zoomed in view of the membrane side coloring the water blue and the lipid phosphate (PO4 beads) red, and take a screenshot.

        3. Analyze the visualization and answer the following questions:
        Q1: Is there any water that penetrates into the membrane phase? (yes/no)
        
        4. Save your work:
        Save the VMD state as "curved-membrane/results/{agent_mode}/curved-membrane.vmd".
        Save the screenshot of the visualization as "curved-membrane/results/{agent_mode}/curved-membrane.png".
        Save the answers to the analysis questions in plain text as "curved-membrane/results/{agent_mode}/answers.txt".
  assert:
    - type: llm-rubric
      subtype: vision
      value: |
          1. Does the visualization show a side view of a bilayer rather than a top-down or oblique view?
          2. Is water present above and below the bilayer, with a mostly empty region in the middle corresponding to the bilayer core?
    - type: llm-rubric
      subtype: text
      value: |
          1. Q1 correct answer: Yes

# Q1 reason: A few water molecules can be seen a little into the membrane.

# 3. Curved membrane trajectory inspection
- vars:
    question: |
        1. Please load the Martini coarse-grained membrane simulation from
        "trajectory-inspection/data/trajectory-inspection.gro" into VMD.

        2. Load the trajectory file
        "trajectory-inspection/data/trajectory-inspection_3to5us.xtc".

        3. Render a titled side view of the membrane using the last frame of the trajectory.

        - Show the PO4 beads of POPC lipids in light brown.
        - Show the PO4 beads of CDL0 lipids in red.
        - Display the simulation box in blue.

        Save the rendered image as:
        "trajectory-inspection/results/{agent_mode}/membrane-curved-tilted-side-5us.jpg"

        4. Perform curvature-based lipid distribution analysis for all the frames in the trajectory and generate the following figures:

        - A 2D density heatmap of CDL0 lipids across the curved membrane surface.
        - A 2D density heatmap of POPC lipids across the curved membrane surface.
        - A plot of lipid relative enrichment versus membrane mean curvature for both POPC and CDL0.

        Save the generated figures as:

        "trajectory-inspection/results/{agent_mode}/avg_2d_dens_CDL0.png"
        "trajectory-inspection/results/{agent_mode}/avg_2d_dens_POPC.png"
        "trajectory-inspection/results/{agent_mode}/relative_enrichment.png"

        5. Analyze the trajectory and answer the following questions:

        Q1: Are there more than 3000 frames in the trajectory? (yes/no)

        Q2: Is the ratio of POPC lipids to the neutral cardiolipin (CDL0) 8:1? (yes/no)

        Q3: Do the CDL0 lipids enrich in the negatively curved membrane regions? (yes/no)

        Q4: Does the total lipid density change significantly with membrane mean curvature? (yes/no)

        Q5: Do the POPC lipids enrich in the negatively curved membrane regions? (yes/no)

        6. Save the answers to the analysis questions in plain text as
        "trajectory-inspection/results/{agent_mode}/answers.txt".

  assert:
    - type: llm-rubric
      subtype: vision
      gs_file: trajectory-inspection/GS/membrane-curved-tilted-side-5us.jpg
      rs_file: trajectory-inspection/results/{agent_mode}/membrane-curved-tilted-side-5us.jpg
      value: |
          1. Do both images show a side view of a curved lipid bilayer rather than a top-down view?
          2. Are two lipid species visible with different colors, corresponding to POPC and CDL0 PO4 beads?
          3. Does the membrane visibly undulate, with one lipid type enriched on the inner curvature and the other on the outer curvature?
          4. Is the simulation box shown in blue with the membrane inside the box

    - type: llm-rubric
      subtype: vision
      gs_file: trajectory-inspection/GS/avg_2d_dens_CDL0.png
      rs_file: trajectory-inspection/results/{agent_mode}/avg_2d_dens_CDL0.png
      value: |
          1. Do both images show a curved membrane band visualized as a heatmap density map?
          2. Is higher CDL0 density concentrated along the inner side of the curved membrane in both images?

    - type: llm-rubric
      subtype: vision
      gs_file: trajectory-inspection/GS/avg_2d_dens_POPC.png
      rs_file: trajectory-inspection/results/{agent_mode}/avg_2d_dens_POPC.png
      value: |
          1. Do both images show a curved membrane band represented as a heatmap density map?
          2. Is higher POPC density concentrated along the outer side of the curved membrane in both images?

    - type: llm-rubric
      subtype: vision
      gs_file: trajectory-inspection/GS/relative_enrichment.png
      rs_file: trajectory-inspection/results/{agent_mode}/relative_enrichment.png
      value: |
          1. Do both plots show enrichment values plotted against membrane mean curvature?
          2. Are two curves corresponding to POPC and CDL0 visible in both plots?
          3. Is a horizontal reference line around enrichment = 1 present in both plots?

    - type: llm-rubric
      subtype: text
      value: |
          1. Q1 correct answer: No (the trajectory contains 2000 frames).
          2. Q2 correct answer: No (the lipid ratio is 9:1).
          3. Q3 correct answer: Yes (CDL0 lipids enrich in negatively curved regions).
          4. Q4 correct answer: Yes (relative density varies roughly between 0.8 and 1.3).
          5. Q5 correct answer: No (POPC lipids enrich in positively curved regions, not negative).