Hello,
I have some questions regarding EcoliTox/RetroPathRL.
My aim is to predict the toxicity of some compounds, and to achieve this, I want to modify RetroPathRL to accept a list of compounds and return their potential toxicity.
I modified 'calculate_organisms.py' and used the 'import_organism_from_csv()' function to input my list of compounds. My list of compounds follows a format similar to the CSV sink files.
However, when I run the program multiple times, the model selected between other_MLP, big_MLP, medium_MLP, and small_MLP varies significantly (with a distribution of 15%/32%/38%/16% over 1000 executions). I'm unsure if the model selection should remain relatively consistent or if such variation is expected. Additionally, the toxicity values fluctuate across executions; for instance, a compound may have a toxicity of -1 in one run, 0 in another, and -0.5 in another.
My question is: is it normal for the model selection to change with each execution? This seems to affect the toxicity values. Also, how should I interpret these toxicity values? For instance, should values between -1 and 0 be considered nearly non-toxic and treated accordingly?
Thank you in advance,
Quentin FERNANDEZ DE GRADO
Hello,
I have some questions regarding EcoliTox/RetroPathRL.
My aim is to predict the toxicity of some compounds, and to achieve this, I want to modify RetroPathRL to accept a list of compounds and return their potential toxicity.
I modified 'calculate_organisms.py' and used the 'import_organism_from_csv()' function to input my list of compounds. My list of compounds follows a format similar to the CSV sink files.
However, when I run the program multiple times, the model selected between other_MLP, big_MLP, medium_MLP, and small_MLP varies significantly (with a distribution of 15%/32%/38%/16% over 1000 executions). I'm unsure if the model selection should remain relatively consistent or if such variation is expected. Additionally, the toxicity values fluctuate across executions; for instance, a compound may have a toxicity of -1 in one run, 0 in another, and -0.5 in another.
My question is: is it normal for the model selection to change with each execution? This seems to affect the toxicity values. Also, how should I interpret these toxicity values? For instance, should values between -1 and 0 be considered nearly non-toxic and treated accordingly?
Thank you in advance,
Quentin FERNANDEZ DE GRADO