feat: add ld50 data (#529)

Author: kjappelbaum

data/tabular/ld50_catmos/meta.yaml

@@ -0,0 +1,145 @@
+---
+name: ld50_catmos
+description: |-
+    Acute toxicity LD50 measures
+    the most conservative dose that can lead to lethal adverse effects.
+    The higher the dose, the more lethal of a drug.
+    We aggregated the data from multiple SMILES by computing the mean.
+targets:
+    - id: CATMoS_LD50_mgkg
+      description: Acute Toxicity LD50.
+      units: mg/kg
+      type: continuous
+      names:
+          - noun: acute oral toxicity rat LD50
+          - noun: acute oral toxicity (LD50 in rats)
+      uris:
+          - http://www.bioassayontology.org/bao#BAO_0002117
+      significant_digits: 1
+    - id: log10_LD50
+      description: Acute Toxicity LD50.
+      units: log10(mg/kg)
+      type: continuous
+      names:
+          - noun: log10 acute oral toxicity rat LD50
+          - noun: log10 acute oral toxicity (LD50 in rats)
+          - noun: log10 LD50 in rats (oral exposure)
+          - noun: log10 rat LD50 (oral exposure)
+      significant_digits: 2
+    - id: num_ghose_violations
+      description: Ghose filter violations
+      type: ordinal
+      significant_digits: 0
+      names:
+          - noun: Ghose filter violations
+          - noun: violations of the Ghose filter
+    - id: num_lead_likeness_violations
+      description: Lead likeness filter violations
+      type: ordinal
+      significant_digits: 0
+      names:
+          - noun: lead likeness filter violations
+          - noun: violations of the lead likeness filter
+    - id: num_lipinski_violations
+      description: Lipinski filter violations
+      type: ordinal
+      significant_digits: 0
+      names:
+          - noun: Lipinski rule violations
+          - noun: violations of the Lipinski rules
+    - id: molecular_mass
+      description: Molecular mass
+      type: continuous
+      units: g/mol
+      names:
+          - noun: molecular mass
+          - noun: molecular weight
+    - id: num_carbon_atoms
+      description: Number of carbon atoms
+      type: ordinal
+      significant_digits: 0
+      names:
+          - noun: carbon atoms
+    - id: num_oxygen_atoms
+      description: Number of oxygen atoms
+      type: ordinal
+      significant_digits: 0
+      names:
+          - noun: oxygen atoms
+identifiers:
+    - id: SMILES
+      type: SMILES
+      description: SMILES
+license: CC BY 4.0
+links:
+    - url: https://ehp.niehs.nih.gov/doi/full/10.1289/EHP8495#supplementary-materials
+      description: corresponding publication
+num_points: 9032
+bibtex:
+    - |-
+      @article{Mansouri_2021, title={CATMoS: Collaborative Acute Toxicity Modeling Suite},
+      volume={129},
+      ISSN={1552-9924},
+      url={http://dx.doi.org/10.1289/EHP8495},
+      DOI={10.1289/ehp8495},
+      number={4},
+      journal={Environmental Health Perspectives},
+      publisher={Environmental Health Perspectives},
+      author={Mansouri, Kamel and Karmaus, Agnes L. and Fitzpatrick, Jeremy
+      and Patlewicz, Grace and Pradeep, Prachi and Alberga, Domenico and
+      Alepee, Nathalie and Allen, Timothy E.H. and Allen, Dave and Alves, Vinicius M.
+      and Andrade, Carolina H. and Auernhammer, Tyler R. and Ballabio, Davide and
+      Bell, Shannon and Benfenati, Emilio and Bhattacharya, Sudin and
+      Bastos, Joyce V. and Boyd, Stephen and Brown, J.B. and Capuzzi, Stephen J. and
+      Chushak, Yaroslav and Ciallella, Heather and Clark, Alex M. and
+      Consonni, Viviana and Daga, Pankaj R. and Ekins, Sean and Farag, Sherif and
+      Fedorov, Maxim and Fourches, Denis and Gadaleta, Domenico and Gao, Feng and
+      Gearhart, Jeffery M. and Goh, Garett and Goodman, Jonathan M. and
+      Grisoni, Francesca and Grulke, Christopher M. and Hartung, Thomas and
+      Hirn, Matthew and Karpov, Pavel and Korotcov, Alexandru and
+      Lavado, Giovanna J. and Lawless, Michael and Li, Xinhao and
+      Luechtefeld, Thomas and Lunghini, Filippo and Mangiatordi, Giuseppe F. and
+      Marcou, Gilles and Marsh, Dan and Martin, Todd and Mauri, Andrea and
+      Muratov, Eugene N. and Myatt, Glenn J. and Nguyen, Dac-Trung and
+      Nicolotti, Orazio and Note, Reine and Pande, Paritosh and
+      Parks, Amanda K. and Peryea, Tyler and Polash, Ahsan H. and
+      Rallo, Robert and Roncaglioni, Alessandra and Rowlands, Craig and
+      Ruiz, Patricia and Russo, Daniel P. and Sayed, Ahmed and Sayre, Risa and
+      Sheils, Timothy and Siegel, Charles and Silva, Arthur C. and Simeonov, Anton and
+      Sosnin, Sergey and Southall, Noel and Strickland, Judy and Tang, Yun and
+      Teppen, Brian and Tetko, Igor V. and Thomas, Dennis and Tkachenko, Valery and
+      Todeschini, Roberto and Toma, Cosimo and Tripodi, Ignacio and
+      Trisciuzzi, Daniela and Tropsha, Alexander and Varnek, Alexandre and
+      Vukovic, Kristijan and Wang, Zhongyu and Wang, Liguo and
+      Waters, Katrina M. and Wedlake, Andrew J. and Wijeyesakere, Sanjeeva J. and
+      Wilson, Dan and Xiao, Zijun and Yang, Hongbin and Zahoranszky-Kohalmi, Gergely and
+      Zakharov, Alexey V. and Zhang, Fagen F. and Zhang, Zhen and Zhao, Tongan and
+      Zhu, Hao and Zorn, Kimberley M. and Casey, Warren and Kleinstreuer, Nicole C.},
+      year={2021}, month=apr }
+templates:
+    - The {#molecule|chemical|compound!} with the {SMILES__description} {#representation of |!}{SMILES#} {#shows|exhibits|displays!} an {CATMoS_LD50_mgkg__names__noun} of {CATMoS_LD50_mgkg#} {CATMoS_LD50_mgkg__units}.
+    - The {#molecule|chemical|compound!} with the {SMILES__description} {#representation of |!}{SMILES#} {#shows|exhibits|displays!} a {log10_LD50__names__noun} of {log10_LD50#} {log10_LD50__units}.
+    - |
+      Task: Determine the acute oral toxicity and molecular properties of a {#molecule|chemical|compound!} given the {SMILES__description}.
+      Input: {SMILES#}
+      Desired Output: {CATMoS_LD50_mgkg__names__noun}, {log10_LD50__names__noun}, {num_ghose_violations__names__noun}, {num_lead_likeness_violations__names__noun}, {num_lipinski_violations__names__noun}, {molecular_mass__names__noun}, {num_carbon_atoms__names__noun}, {num_oxygen_atoms__names__noun}
+      Output: {CATMoS_LD50_mgkg#} {CATMoS_LD50_mgkg__units}, {log10_LD50#} {log10_LD50__units}, {num_ghose_violations#}, {num_lead_likeness_violations#}, {num_lipinski_violations#}, {molecular_mass#} {molecular_mass__units}, {num_carbon_atoms#}, {num_oxygen_atoms#}
+    - |
+      Context: You are {#an assistant|researcher|scientist!} in a pharmaceutical company. Your {#boss|superior|department head!} has asked you to {#design|create|synthesize!} a new drug.
+      User: The {#drug|compound|chemical!} should have a {CATMoS_LD50_mgkg__names__noun} of {CATMoS_LD50_mgkg#} {CATMoS_LD50_mgkg__units}, {num_ghose_violations#} {num_ghose_violations__names__noun}, {num_lead_likeness_violations#} {num_lead_likeness_violations__names__noun}, {num_lipinski_violations#} {num_lipinski_violations__names__noun}, {molecular_mass#} {molecular_mass__names__noun} {molecular_mass__units}, {num_carbon_atoms#} {num_carbon_atoms__names__noun}, and {num_oxygen_atoms#} {num_oxygen_atoms__names__noun}.
+      Assistant: {#Happy to help!|Sure!|Of course!} The {#molecule|chemical|compound!} with the {SMILES__description} {#representation of |!}{SMILES#} {#shows|exhibits|displays!} the desired properties.
+    - |
+      User: I need a {#drug|compound|chemical!} with a {log10_LD50__names__noun} of {log10_LD50#} {log10_LD50__units}.
+      Assistant: {#Happy to help!|Sure!|Of course!} Can you provide me with more {#constraints|details|information!}?
+      User: The {#drug|compound|chemical!} should have {num_ghose_violations#} {num_ghose_violations__names__noun}, {num_lead_likeness_violations#} {num_lead_likeness_violations__names__noun}, {num_lipinski_violations#} {num_lipinski_violations__names__noun}, {num_carbon_atoms#} {num_carbon_atoms__names__noun}, and {num_oxygen_atoms#} {num_oxygen_atoms__names__noun}.
+      Assistant: The {#molecule|chemical|compound!} with the {SMILES__description} {#representation of |!}{SMILES#} {#shows|exhibits|displays!} the desired properties.
+    - |
+      User: I need a {#drug|compound|chemical!} with a {CATMoS_LD50_mgkg__names__noun} of {CATMoS_LD50_mgkg#} {CATMoS_LD50_mgkg__units}.
+      Assistant: {#Happy to help!|Sure!|Of course!} Can you provide me with more {#constraints|details|information!}?
+      User: The {#drug|compound|chemical!} should have a {num_carbon_atoms#} {num_carbon_atoms__names__noun}, {num_oxygen_atoms#} {num_oxygen_atoms__names__noun}, and a {molecular_mass__names__noun} of {molecular_mass#} {molecular_mass__units}. Could you please only provide me with the {SMILES__description} and return no other information?
+      Assistant: {SMILES#}
+    - |
+      User: I am looking for a {#drug|compound|chemical!} with a {log10_LD50__names__noun} of {log10_LD50#} {log10_LD50__units}.
+      Assistant: {#That's interesting!|Interesting!|I see!} Can you provide me with more {#constraints|details|information!}?
+      User: The {#drug|compound|chemical!} should have {num_ghose_violations#} {num_ghose_violations__names__noun}, {num_lead_likeness_violations#} {num_lead_likeness_violations__names__noun}, {num_lipinski_violations#} {num_lipinski_violations__names__noun}, {num_carbon_atoms#} {num_carbon_atoms__names__noun}, and {num_oxygen_atoms#} {num_oxygen_atoms__names__noun}. Please return only the {SMILES__description} wrapped as follows [ANSWER]<SMILES>[/ANSWER].
+      Assistant: [ANSWER]{SMILES#}[/ANSWER]

data/tabular/ld50_catmos/transform.py

@@ -0,0 +1,36 @@
+import pandas as pd
+from huggingface_hub import hf_hub_download
+
+
+def process():
+    file = hf_hub_download(
+        repo_id="kjappelbaum/chemnlp-ld50catmos",
+        filename="cleaned_ld50.csv",
+        repo_type="dataset",
+    )
+    df = pd.read_csv(file)
+    print(len(df))
+    df[
+        [
+            "num_ghose_violations",
+            "num_lead_likeness_violations",
+            "num_lipinski_violations",
+            "num_carbon_atoms",
+            "num_oxygen_atoms",
+        ]
+    ] = df[
+        [
+            "num_ghose_violations",
+            "num_lead_likeness_violations",
+            "num_lipinski_violations",
+            "num_carbon_atoms",
+            "num_oxygen_atoms",
+        ]
+    ].astype(
+        int
+    )
+    df.to_csv("data_clean.csv", index=False)
+
+
+if __name__ == "__main__":
+    process()