Hi Fabian Ugbe

I have used Open3DQSAR software in my project. a list of commads which you need are:

#The following command imports a SDF dataset

import type=sdf file=ref_e2.sdf

#The following removes from the dataset

#the specified compound numbers, i.e.

#1, 6, 8, 13, 14, 16, 17, 20, 21, 27, 30,

#31, 32, 33, 34, 35, 39, 52, 53, 54

remove_object object_list=1,6,8,13,14,16,17,20,21,27,30,31,32,33,34,35,39,52,53,54

#The following command sets a grid box with a 2.0

#angstrom step size and a 5.0 angstrom outgap

#around the molecules

box step=2.0

#The following command computes a steric field

calc_field type=vdw

#The following command computes an ESP field

calc_field type=mm_ele

#The following commands set min/max cutoffs on the VdW field

cutoff type=max level=30.0

cutoff type=min level=-30.0

#The following command imports a text file with activities

#the first line must be a label with the variable name

#(e.g., affinity, or activity)

#a multi-column file can be read

#to perform regression over multiple

#dependent variables

import type=dependent file=binding_data_36_compounds.txt

#Saves the file after having imported

#both independent and dependent variables

save file=binding.dat

#The following zeroes both negative and positive values

#below a threshold of 0.05 for both probes

zero type=all level=0.05

#The following extracts compounds having ID

#7, 10, 15, 25, 37, 40, 42, 46, 55

#to use them as an external test set

set id_list=7,10,15,25,37,40,42,46,55 attribute=testset

#The following removes X variables with a SD

#below a threshold of 0.1

sdcut level=2.0

#The following removes n-level variables

#of 2nd, 3rd and 4th level from both fields

nlevel

remove_x_vars type=nlevel

#The following command performs

#Block Unscaled Weighting

scale_x_vars type=buw

#The following command performs

#PLS analysis extracting 5 PCs

pls pc=5 file=pls_results.sdf

#The following command performs

#external prediction on the 9-compound

#test set previously extracted using 5 PCs

predict pc=5 file=ext_pred_results.sdf

#The following commands performs

#leave-one-out, leave-two-out,

#leave-many-out cross-validation

#with 5 PCs

cv pc=5 type=loo file=loo_results.sdf

cv pc=5 type=lto file=lto_results.sdf

cv pc=5 type=lmo runs=20 groups=5 file=lmo_results.sdf

#The following commands performs

#Smart Region Definition (SRD) variable grouping

srd pc=5 collapse=yes critical_distance=1.0 \

collapse_distance=2.0 type=weights

#Now we save an intermediate file

#after SRD

save file=binding_after_srd.dat

#The following commands performs

#FFD variable selection

ffdsel pc=5 type=lmo runs=20 \

percent_dummies=20 use_srd_groups=yes \

combination_variable_ratio=1.0 fold_over=no

#Now we save an intermediate file

#after FFD variable selection

save file=binding_after_ffdsel.dat

#The following commands removes

#variable selected as unimportant

#by FFDSEL

remove_x_vars type=ffdsel

#Now let's recalculate a model

#after variable reduction

pls pc=5

#And let's challenge its performance

#through LOO, LTO, LMO cross-validation

cv pc=5 type=loo

cv pc=5 type=lto

cv pc=5 type=lmo groups=5 runs=20

#And also through external prediction

predict pc=5

#Now we save a grid file

#with PLS pseudo-coefficients

#for visualization in MOE, PyMOL

#and Maestro

export type=coefficients pc=3 file=ffdsel_coefficients \

format=insight interpolate=3

export type=coefficients pc=3 file=ffdsel_coefficients \

format=moe interpolate=3

export type=coefficients pc=3 file=ffdsel_coefficients \

format=maestro interpolate=3

#Now we reload the data file

#before we carried out FFDSEL

load file=binding_after_srd.dat

pls pc=5

#And we attempt a different variable selection

#procedure, namely IVE-PLS

uvepls pc=5 type=lmo \

use_srd_groups=no \

uve_alpha=90 uve_m=y ive=y \

save_ram=n ive_percent_limit=50

#Now we save an intermediate file

#after IVE-PLS variable selection

save file=binding_after_uvepls.dat

remove_x_vars type=uvepls

#Now let's recalculate a model

#after variable reduction

pls pc=5

#And let's challenge its performance

#through LOO, LTO, LMO cross-validation

cv pc=5 type=loo

cv pc=5 type=lto

cv pc=5 type=lmo groups=5 runs=20

#And also through external prediction

predict pc=5

#Now we save a grid file

#with PLS pseudo-coefficients

#for visualization in MOE, PyMOL

#and Maestro

export type=coefficients pc=3 file=uvepls_coefficients \

format=insight interpolate=3

export type=coefficients pc=3 file=uvepls_coefficients \

format=moe interpolate=3

export type=coefficients pc=3 file=uvepls_coefficients \

format=maestro interpolate=3

Open attached file with notepad software.

I you still have a problem, we can have a video chat in google meet. You can contact me via following email:

[email protected]

Dear Dr. Bakhtyar Sepehri,

Thank you so much for adding a well detailed answer. I think the procedure you provided dwells on building the QSAR model.

I was able to build a 3D-QSAR model earlier. I was equally able to design a set of new compounds. Since these new compounds were only designed theoretically, experimental activities not applicable. My question now is how can I predict the activities of these new compounds with the previously built model?