Pepinvent prior and sample input TOML file

[ms-yk]

I would appreciate it very much if you provide Pepinvent prior and sample input TOML file.

[halx]

Hi,

many thanks for your interest in REINVENT and welcome to the community!

As for the prior I would need an upload directory as I am otherwise unable to make the file available in a sensible manner. The configuration is rather straightforware as it is standard Reinforcement Learning and the input SMILES are provided in CHUCKLES representation.

# One masked peptide with CHUCKLES representation per line
# ? for mask

?|N[C@@H](CO)C(=O)|?|N[C@@H](Cc1ccc(O)cc1)C(=O)|N(C)[C@@H]([C@@H](C)O)C(=O)|N[C@H](Cc1c[nH]cn1)C(=O)|N[C@@H](CC(=O)N)C2(=O)

run_type = "staged_learning"
device = "cuda:0"
use_cuda = true
tb_logdir = "tb_logs"
json_out_config = "pepinvent_staged_learning.json"


[parameters]
prior_file = "pepinvent.prior"
agent_file = "pepinvent.prior"
summary_csv_prefix = "pepinvent"
use_checkpoint = false
purge_memories = false
smiles_file = "pepinvent.smi"  # see above
sample_strategy = "multinomial"
distance_threshold = 100
batch_size = 64
randomize_smiles = true
unique_sequences = true
temperature = 1.0

[learning_strategy]
type = "dap"
sigma = 128
rate = 0.0001

[diversity_filter]
type = "IdenticalMurckoScaffold"
bucket_size = 25
minscore = 0.4
minsimilarity = 0.4
penalty_multiplier = 0.5

[[stage]]
max_steps = 100
max_score = 0.6
chkpt_file = "test1.chkpt"
termination = "simple"
min_steps = 25

[stage.scoring]
type = "geometric_mean"
[[stage.scoring.component]]

[stage.scoring.component.MolecularWeight]
[[stage.scoring.component.MolecularWeight.endpoint]]
name = "Molecular weight"
weight = 1

[stage.scoring.component.MolecularWeight.endpoint.transform]
type = "double_sigmoid"
high = 1500.0
low = 400.0
coef_div = 500.0
coef_si = 20.0
coef_se = 20.0

The real issue, however, is to have meaningful scoring components in place. We do not provide much of those at the moment.

Many thanks,
Hannes.

[ms-yk]

Hi Hannes,

Thank you so much for your quick reply. Is there a way to create a pepinvent prior model from current shared REINVENT4 sources?

Best regards,
Yoshirou

[jiazhenhe]

There is no ready-to-use script for creating pepinvent prior currently. But one can adapt the script for creating and training Mol2Mol to pepinvent.

[ms-yk]

Thank you for the information. I understand.