LunarRegolith

This repository contains details on the reanalysis of lunar-grown plant data performed for a NASA GeneLab project, conducted in collaboration between the Causal Inference and Lunar Regolith subgroups.

Below, you can find a step-by-step instruction on what I've done so far.

This reanalysis pertains to the OSD-476 study available via the NASA Open Science Data Repository (OSDR). The data used in this reanalysis consists of four images grown in lunar regolith simulant and lunar regolith.

The goal of this study is to identify the genes that have a causal relationship with the change in the phenotype of plants grown on lunar regolith from Apollo 11, 12, and 17 missions in comparison to those grown on lunar regolith simulant.

To properly proceed with the next step, each image must contain only one plant. Since the images available in OSDR have seven plants each, I had to manually remove all but one plant for each photo. To achieve this, I used Photoroom's AI retouch feature. This way, I ended up with 4 × 7 = 28 photos. You can see the original and edited images in SOAPP/images.

To perform a detailed analysis of the sizes and color compositions of the plants, I used SOAPP. You can find detailed instructions on how to run SOAPP here. It's easy, don't be afraid! I used Docker. I chose custom settings for each image.

I saved all the outputs in two csv files. You can find them in the SOAPP/output directory. Here's an explanation of what is in them:

• Results.csv contains all the data regarding the size of the plants.
• Color_results.csv contains all the data regarding the color of the plants.

SOAPP/output/images_to_sequencing.csv is the file that matches SOAPP output with differential gene expression data, i.e., the GSM numbers.

How did I match them?

I downloaded the metadata zip file (OSD-476_metadata_OSD-476-ISA.zip) from the OSDR repository. The s_OSD-476.txt file allowed me to associate the GSM numbers with their corresponding lunar regolith type (JSC-1A, Apollo 11, Apollo 12, and Apollo 17). This enabled me to link the Apollo regolith samples directly to their GSM identifiers, but I still didn't know which JSC-1A plant was sequenced in each case. To resolve this, I studied Figure 3 from the associated paper, and i_Investigation.txt metadata file (from OSD-476_metadata_OSD-476-ISA.zip). That file includes the note:

"In Figure 3 [...] one can see the full designation of each plant used in the transcriptomes, so “P1J4JSC1A” means Plate 1, JSC plant 4, (JSC spelled out just to clarify for figure) and correlates to ‘Col-0, Leaves, JSC-1A, replicate 1’ in the transcriptome data set."

It also states:

"Although there were four representative JSC1A plants on each plate, only one of those plants (the plant in position 4 in each plate) was used from each plate for the transcriptome analyses."

Based on this information, I inferred that the JSC-1A 4 plants were sequenced in each case (check SOAPP/images/original_images/LSDS-12_morphometric-photography_Lunar_plate_map.jpg or below image for reference).

Why are there only 16 sequences available when 28 plants were grown?

Only one plant per regolith type was sequenced on each of the four plates, resulting in:

• 4 × JSC-1A (position 4)

• 4 × Apollo 11

• 4 × Apollo 12

• 4 × Apollo 17

  = 16 sequenced samples

Note: An additional four samples associated with JSC-1A-grown plants were sequenced (GSM5691027, GSM5691028, GSM5691029, GSM5691030), but their exact plant origin could not be confidently determined from the available metadata.

You can run the following line to generate the combined file:

python 1_combine_and_sum.py

The output file is in RNAseq/combined_with_rna_sums.csv. Note that RNAseq data is only available for 16 plants and therefore only these 16 are included in this file.

This is some nonsense of me testing

# 1) Fresh env with an old-enough Python
CONDA_SUBDIR=osx-64 conda create -n crisp-py37 python=3.7 -y
conda activate crisp-py37
conda config --env --set subdir osx-64

# 2) Make pip/setuptools/wheel compatible with the era
pip install "pip==20.3.4" "setuptools==58.5.3" "wheel==0.38.4"

# 3) Pre-install the hard compiled deps via conda (avoids building numpy)
conda install -y -c conda-forge "numpy==1.17.3" "bottleneck==1.3.2" "scipy==1.5.*" "pandas==1.1.*"

# 4) Now install the repo's requirements without trying to replace those deps
cd crispv1.1
python -c "import platform; print(platform.machine())"   # should print x86_64

# conda way (recommended)
CONDA_SUBDIR=osx-64 conda install -n crisp-py37 -c conda-forge "tbb=2020.2" -y

# OR pip fallback if you prefer
pip install "tbb==2021.6.0"

# create a copy of requirements without the TBB line
grep -v -i '^tbb==' requirements.txt > requirements.no-tbb.txt

# now continue your approach (no deps, since we preinstalled the tricky ones)
pip install --no-deps -r requirements.no-tbb.txt

After that run the code to create the CRISP input files

python make_crisp_input.py

conda activate crisp-py37

pip install \
  "attrs==21.4.0" \
  "importlib-metadata==4.13.0" \
  "packaging==21.3" \
  "python-dateutil==2.8.2" \
  "requests==2.28.2" \
  "typing-extensions==4.1.1" \
  "protobuf<4" \
  "watchdog==2.1.9" \
  toml

pip install decorator==5.1.1

pip install "jinja2<3.1"

pip install "click<8.1" "blinker==1.4" "pyarrow==5.0.0" "jsonschema<4.0" "toolz==0.11.2"

pip install "networkx==2.5"

pip install \
  "networkx==2.5" \
  "matplotlib==3.3.4" \
  "scikit-learn==0.24.2" \
  "statsmodels==0.12.2" \
  "pandas==1.2.5" \
  "seaborn==0.11.2"

pip install "cryptography<37" "pyasn1==0.4.8" "google-cloud-storage==1.42.3"

python 1_make_crisp_input.py \
  --soapp_output "/Volumes/T7/2_Second_iteration_separately/LunarRegolith/SOAPP/output" \
  --crisp_input  "/Volumes/T7/2_Second_iteration_separately/LunarRegolith/CRISP/CRISPv1/input"

python 2_create_crisp_jsons.py

./3_train_all.sh