ChemFH

Overview

What is ChemFH?

One of the major challenges in early drug discovery is the detection of frequent false-positive results (Fig.1), which have seriously interfered with hit identification, thus leading to a waste of time and resource. To avoid the attritions caused by false positive results, ChemFH (Chemical Frequent Hitter), an integrated online platform is developed for the screening and prediction of potential frequent hitters, covering colloidal aggregate, firefly luciferase reporter enzyme inhibition, fluorescence, chemical reactivity and promiscuity. Based on the collection of a large and high-quality database and the application of Graph Neural Networks architectures (Fig.2), ChemFH is able to accomplish reliable detection capability of frequent hitters, thus improving the efficiency of drug R&D.

Figure 1. The distributions of true positive result and different mechanism of false positive results in HTS assay

Figure 2. The main modules and functions of ChemFH

Introduction

Evaluation Mode

The Evaluation Mode allows the input of single molecule which aims to confirm the authenticity of biological campaign screening result. Based on the combination of credible prediction models and useful substructure rules, it is believed that through Evaluation Mode, it is able to comprehensively evaluate the risk and potential of the queried molecule.

Screening Mode

The Screening Mode is able to assist hit identification by screening molecular dataset prior to detect potential frequent hitters. Such function makes it possible to pre-screen molecules with undesirable interfering features, thus increasing the efficiency of drug design, improving the credibility of experimental result and decreasing unnecessary cost.

Performance of Prediction Models

Category	Data			Model (AUC)
Category	Positive	Negative	Total	GCN	GAT	AttentiveFP
Agg	16783	35114	51897	0.937	0.937	0.937
Blue FLuo	4871	38659	43530	0.936	0.938	0.926
GreenFLuo	8319	31352	39671	0.728	0.723	0.720
Other assay interference	1544	2159	3703	0.779	0.777	0.743
FLuc inhibitor	12703	121345	133418	0.969	0.967	0.910
Reactive compound	820	162069	162889	0.992	0.993	0.909
Promiscuous compound	7518	334171	341689	0.921	0.930	0.905

Performance of Mechanism Substructure Rules

Category	Number of substructures	Flagged molecules	Flagged FHs	Accuracy
Agg	26	2651	2265	0.854
Blue FLuo	26	1208	1172	0.970
GreenFLuo	16	274	260	0.949
Other assay interference	7	209	200	0.957
FLuc inhibitor	18	1638	1448	0.884
Reactive compound	3	30	23	0.767
Promiscuous compound	6	733	704	0.960

Information about Other FH Screening Rules

Name	Number of substructures	Screening Scope	Source
PAINS	480	Alpha-screen artifacts and reactive compound	J Med Chem 2010;53:2719–40
BMS	176	Undesirable, reactive compounds	J Chem Inf Model 2006;46:1060–8
GST/GSH FH filter	34	GST/GSH FHs	J Biomol Screen 2016;21:596–607
His-tagged protein FH filter	19	Ni2+ chelators	J Biomol Screen 2014;19:715–26
ALARM NMR	75	Thiol reactive compounds	J Am Chem Soc 2005;127:217–24
Luciferase inhibitor Rule	3	FLuc inhibitors	J Chem Inf Model 2018;58:933–42
Chelator Rule	55	Chelators	ChemMedChem 2010;5:195–9
NTD	105	Unwanted groups, reactive groups and possible HTS interferences	ChemMedChem 2008;3:435–44
Potential electrophilic Rule	119	Reactive compounds	J Chem Inf Model 2012;52:2310–6

Development Environment

The Development Environment of ChemFH

Library	Version
RDkit	2019.03.1
Django	2.2
DGL	0.5.2
DGL-LifeSci	0.2.5
Pytorch	1.6.0
Torchvision	0.7.0
pyecharts	1.8.1

Evaluation Mode

Input Types

The Evaluation Mode allows the input of single molecule which aims to confirm the authenticity of biological campaign screening result. Two input types are provided:

1. By inputting single SMILES string;
2. By drawing molecule from editor below.

Single SMILES string

Step1: Access the Services page via Services-> Evaluation Mode in the navigation bar.

Step2: Select first entry for: Input a single strings and Enter the SMILES string in the SMILES input box.

Step3: Select related evaluation methods for following application. Two types of evaluation methods are provided: specific mechanism FH detection and collected credible FH screening rules application. The prior type includes both prediction models and substructure rules, which are constructed based on the special dataset collection.

Mechanism:

1.Colloidal aggregators: Compound aggregation tends to start when the concentration is above the CAC and end as aggregators form with radius of approximately 30-600 nm. The resulting colloidal aggregators would non-specifically bind to the surface of proteins, thus inducing local protein unfolding, which usually results in destabilization or denaturation of enzymes
2.FLuc inhibitors: Due to its unique catalysis mechanism, FLuc is widely used in a variety of HTS bioluminescence assays, especially in the assay which aims to study gene expression at the transcriptional level. However, the inhibition of Fluc by unexpected FLuc inhibitors would produce interference to HTS assays.
3.Blue/Green fluorescence: Fluorescence is the process by which a molecule, called fluorophore or fluorescent dye, absorbs a photon of light, exciting an electron to a higher energy state. Fluorophores have many applications, including as enzyme substrates, labels for biomolecules, cellular stains and environmental indicators. However, the appearance of fluorescent compound would produce interference to related HTS assays.
4.Reactive compounds: Chemical reactive compounds typically result in the chemical modification of reactive protein residues or, less frequently, the modification of nucleophilic assay reagents.
5.Promiscuous compounds: Promiscuous compounds refer to compounds that specifically bind to different macromolecular targets. These multiple interactions may include unintended targets, thus triggering adverse reactions and other safety issues.
6.Other assay interferences: Alpha-screen, FRET, TR-FRET, absorbance artifacts are included.

Other FH Screening Rules:

1.PAINS: frequent hitters, Alpha-screen artifacts and reactive compound; 480 substructures (J Med Chem 2010;53:2719–40)
2.BMS: undesirable, reactive compounds; 176 substructures (J Chem Inf Model 2006;46:1060–8)
3.GST/GSH FH filter: GST/GSH FHs; 34 substructures ( J Biomol Screen 2016;21:596–607)
4.His-tagged protein FH filter: Ni2+ chelators; 19 substructures (J Biomol Screen 2014;19:715–26)
5.ALARM NMR: thiol reactive compounds; 75 substructures (J Am Chem Soc 2005;127:217–24)
6.Luciferase inhibitor Rule: FLuc inhibitors; 3 substructures (J Chem Inf Model 2018;58:933–42)
7.Chelator Rule: chelators; 55 substructures (ChemMedChem 2010;5:195–9)
8.NTD: unwanted groups, reactive groups and possible HTS interferences; 105 substructures (ChemMedChem 2008;3:435–44)
9.Potential electrophilic Rule: reactive compounds; 119 substructures (J Chem Inf Model 2012;52:2310–6)

Step4: Submit and get results.

Output Types

After submission, the backend will input the queried molecule into the selected models and (or) substructure rules. This page will show a brief overview of the results obtained, including:

1. Visualization block:

1) The status of the queried molecule;
2) The 2D, 3D structure graph of the queried molecule;
3) The radar chart is provided for the better comprehension of the different mechanism prediction result, where Blue Label indicates that this type of mechanism is chosen for prediction, while Black Label indicates that this this type of mechanism is not chosen for prediction.

2.Model Predictions and Substructure Screening blocks:

1) Model Predictions block provide the information about the prediction mechanism and corresponding GCN, GAT, AttentiveFP and average prediction scores;
2) Substructure Screening block provide the information about the prediction mechanism and corresponding flagged substructure name and graph.

3.File Download:

The csv (pdf) file shows the score the accepted status and fragment information for each mechanism for each molecule (in this case, only one molecule), including SMILES, status, GCN score, GAT score, AttentiveFP score, average score, and SMARTS for each rule.

Molecular Editor

Step1: Access the Services page via Services-> Evaluation Mode in the navigation bar.

Step2: Select second entry for: Drawing molecules through the molecule editor and draw a molecule.

Mechanism:

1.Colloidal aggregators: Compound aggregation tends to start when the concentration is above the CAC and end as aggregators form with radius of approximately 30-600 nm. The resulting colloidal aggregators would non-specifically bind to the surface of proteins, thus inducing local protein unfolding, which usually results in destabilization or denaturation of enzymes
2.FLuc inhibitors: Due to its unique catalysis mechanism, FLuc is widely used in a variety of HTS bioluminescence assays, especially in the assay which aims to study gene expression at the transcriptional level. However, the inhibition of Fluc by unexpected FLuc inhibitors would produce interference to HTS assays.
3.Blue/Green fluorescence: Fluorescence is the process by which a molecule, called fluorophore or fluorescent dye, absorbs a photon of light, exciting an electron to a higher energy state. Fluorophores have many applications, including as enzyme substrates, labels for biomolecules, cellular stains and environmental indicators. However, the appearance of fluorescent compound would produce interference to related HTS assays.
4.Reactive compounds: Chemical reactive compounds typically result in the chemical modification of reactive protein residues or, less frequently, the modification of nucleophilic assay reagents.
5.Promiscuous compounds: Promiscuous compounds refer to compounds that specifically bind to different macromolecular targets. These multiple interactions may include unintended targets, thus triggering adverse reactions and other safety issues.
6.Other assay interferences: Alpha-screen, FRET, TR-FRET, absorbance artifacts are included.

Other FH Screening Rules:

1.PAINS: frequent hitters, Alpha-screen artifacts and reactive compound; 480 substructures (J Med Chem 2010;53:2719–40)
2.BMS: undesirable, reactive compounds; 176 substructures (J Chem Inf Model 2006;46:1060–8)
3.GST/GSH FH filter: GST/GSH FHs; 34 substructures ( J Biomol Screen 2016;21:596–607)
4.His-tagged protein FH filter: Ni2+ chelators; 19 substructures (J Biomol Screen 2014;19:715–26)
5.ALARM NMR: thiol reactive compounds; 75 substructures (J Am Chem Soc 2005;127:217–24)
6.Luciferase inhibitor Rule: FLuc inhibitors; 3 substructures (J Chem Inf Model 2018;58:933–42)
7.Chelator Rule: chelators; 55 substructures (ChemMedChem 2010;5:195–9)
8.NTD: unwanted groups, reactive groups and possible HTS interferences; 105 substructures (ChemMedChem 2008;3:435–44)
9.Potential electrophilic Rule: reactive compounds; 119 substructures (J Chem Inf Model 2012;52:2310–6)

Step4: Submit and get results.

Output Types

After submission, the backend will input the queried molecule into the selected models and (or) substructure rules. This page will show a brief overview of the results obtained, including:

1. Visualization block:

1) The status of the queried molecule;
2) The 2D, 3D structure graph of the queried molecule;
3) The radar chart is provided for the better comprehension of the different mechanism prediction result, where Blue Label indicates that this type of mechanism is chosen for prediction, while Black Label indicates that this this type of mechanism is not chosen for prediction.

2.Model Predictions and Substructure Screening blocks:

1) Model Predictions block provide the information about the prediction mechanism and corresponding GCN, GAT, AttentiveFP and average prediction scores;
2) Substructure Screening block provide the information about the prediction mechanism and corresponding flagged substructure name and graph.

3.File Download:

Screening Mode

Input Types

The Screening Mode is able to assist hit identification by screening molecular dataset prior to detect potential frequent hitters. User can upload file (.txt/.sdf) for following screening.

Molecular File

Step1: Access the Services page via Services-> Screening Mode in the navigation bar.

Step2: Select the entry for: Molecule File and select the upload file (.txt/.sdf) for following screening.

Mechanism:

1.Colloidal aggregators: Compound aggregation tends to start when the concentration is above the CAC and end as aggregators form with radius of approximately 30-600 nm. The resulting colloidal aggregators would non-specifically bind to the surface of proteins, thus inducing local protein unfolding, which usually results in destabilization or denaturation of enzymes
2.FLuc inhibitors: Due to its unique catalysis mechanism, FLuc is widely used in a variety of HTS bioluminescence assays, especially in the assay which aims to study gene expression at the transcriptional level. However, the inhibition of Fluc by unexpected FLuc inhibitors would produce interference to HTS assays.
3.Blue/Green fluorescence: Fluorescence is the process by which a molecule, called fluorophore or fluorescent dye, absorbs a photon of light, exciting an electron to a higher energy state. Fluorophores have many applications, including as enzyme substrates, labels for biomolecules, cellular stains and environmental indicators. However, the appearance of fluorescent compound would produce interference to related HTS assays.
4.Reactive compounds: Chemical reactive compounds typically result in the chemical modification of reactive protein residues or, less frequently, the modification of nucleophilic assay reagents.
5.Promiscuous compounds: Promiscuous compounds refer to compounds that specifically bind to different macromolecular targets. These multiple interactions may include unintended targets, thus triggering adverse reactions and other safety issues.
6.Other assay interferences: Alpha-screen, FRET, TR-FRET, absorbance artifacts are included.

Other FH Screening Rules:

1.PAINS: frequent hitters, Alpha-screen artifacts and reactive compound; 480 substructures (J Med Chem 2010;53:2719–40)
2.BMS: undesirable, reactive compounds; 176 substructures (J Chem Inf Model 2006;46:1060–8)
3.GST/GSH FH filter: GST/GSH FHs; 34 substructures ( J Biomol Screen 2016;21:596–607)
4.His-tagged protein FH filter: Ni2+ chelators; 19 substructures (J Biomol Screen 2014;19:715–26)
5.ALARM NMR: thiol reactive compounds; 75 substructures (J Am Chem Soc 2005;127:217–24)
6.Luciferase inhibitor Rule: FLuc inhibitors; 3 substructures (J Chem Inf Model 2018;58:933–42)
7.Chelator Rule: chelators; 55 substructures (ChemMedChem 2010;5:195–9)
8.NTD: unwanted groups, reactive groups and possible HTS interferences; 105 substructures (ChemMedChem 2008;3:435–44)
9.Potential electrophilic Rule: reactive compounds; 119 substructures (J Chem Inf Model 2012;52:2310–6)

Step4: Submit and get results.

Output Types

After submission, the backend will input the queried molecule into the selected models and (or) substructure rules. This page will show a brief overview of the results obtained, including:

1. Summary block:

1) Molecules entry indicates the total number of input SMILES strings;
2) Mechanism entry indicates the types of chosen screening mechanism;
3) Other rules entry indicates types of chosen screening structural rules;
4) Link entry provided the download service about the detailed screening result;
5) Pie chart is provided for the better comprehension of the prediction status of queried molecules.

2. Result block:

1) INDEX entry indicates the index of the input SMILES;
2) STATUS entry indicates the final screening result: Accepted (low FH risk), Intermediate (potential FH risk) and Rejected (high FH risk);
3) STRUCTURE entry is the molecular image;
4) SMILES entry is the input SMILES string;
5) Specific evaluation information for each molecule can be viewed by clicking on the VIEW button under the DETAIL entry.

3. File Download:

4. Specific Result:

Click the VIEW button to see specific information about each molecule.

Navigation

Overview

Introduction

Development Environment

Evaluation Mode

Screening Mode