Predictive Modeling

A key component in the Q-MAP platform is its predictive models. These are supervised learning models trained, typically as classifiers, to identify a given activity based on a molecular structure. These activities range across target ligand-protein bindings, ADME properties, toxicities, and side-effects. The information below provides further information on our suite of predictive models.

All models were evaluated on data that was held back from the learning method used.

ModelBalanced AccuracySpecificitySensitivityAUC
Absorption, Distribution, Metabolism, Execretion,
and Other Toxicities (ADMET)
Blood Brain Barrier Permeability (BBP)
PLS – 3D – SDAR
0.7350.7000.7690.823
BBBP
PLS – ECFP
0.8190.9000.7880.895
Cardiotoxicity – hERG Inhibition0.8450.9010.7880.915
Cardiotoxicity – QT Prolongation (QTP)0.7340.7860.6820.727
Mutagenicity – TA980.7800.7270.8330.851
Mutagenicity – TA1000.7100.7110.7080.721
Nephrotoxicity – Interstitial Nephritis
(IN)
0.8710.9230.8610.932
Receptor Binding and Functionality
5-HT2A – Agonist vs Antagonist0.7590.7930.7250.855
5-HT2A – Ketanserin – Binding Affinity0.8500.9570.7060.818
5-HT2A – Functionality0.8000.7500.8750.969
5-HT7 – LSD – Binding Affinity0.8440.8650.7500.885
Dopamine (D2) – Methylspiperone –
Binding Affinity
0.9091.0000.8330.867
Dopamine (D2L) – Binding Affinity0.7500.8330.7310.776
Dopamine (D2) – Functionality0.8240.7080.9390.869
Dopamine (D3) – [3H] – Spiperone –
Binding Affinity
0.8370.9230.8000.926
Opioid Reversal – Delta Receptor –
Binding Affinity
0.8480.8070.8880.921
Opioid Reversal – Delta Receptor –
Functionality
0.9171.0000.8330.931
Opioid Reversal – Kappa Receptor –
Binding Affinity
0.8040.8570.7500.866
Opioid Reversal – Kappa Receptor –
Functionality
0.8180.8670.7690.913
Opioid Reversal – Mu Receptor –
Binding Affinity
0.8220.8190.8240.890
Opioid Reversal – Mu Receptor –
Functionality
0.8140.7330.8950.912

ADMET Modeling

ADMET Modeling Description: This set of models serves to address drug properties aligning with Absorption, Distribution, Metabolism, Excretion, and General Toxicities.

Blood Brain Barrier Permeability (BBBP)

Description: The Blood Brain Barrier (BBB) is formed by the endothelium of brain micro vessels and provides numerous important functions. Functions include the delivery of oxygen and nutrients, the removal of waste products, and provide protection against potentially toxic agents and pathogens. Depending on the type of drug being synthesized, transportation across the BBB can play a crucial role in the success of efficacy.

Classification = 1 –> Molecule Does Pass BBB

Classification = 0 –> Molecule Does NOT Pass BBB

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR MetricTraining ScoreValidation ScoreTest Score
Balance Accuracy0.9500.7220.735
Specificity 0.9630.7000.700
Sensitivity0.9370.7430.769
Area Under the Curve (AUC)0.9950.7570.823
PLS – ECFP – MetricTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9760.7490.819
Specificity0.9880.7620.900
Sensitivity0.9710.7440.788
Area Under the Curve (AUC)0.9980.8270.895

Cardiotoxicity: hERG Inhibition

Description: Drug-induced hERG inhibition potentially leads to long QT syndrome and is part of the cardiac safety profile for pre-clinical drug development issued by the FDA. There are many in silico models for predicting with some built into commercially available software packages. Spektron is including this model to provide modeling to cover major safety considerations in its suite of models.

Classification = 1 –> Molecule Indicates hERG Activity

Classification = 0 –> Molecule Does Not Indicate hERG Activity

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.8760.8450.845
Specificity0.9250.9010.901
Sensitivity0.8270.7880.788
Area Under the Curve (AUC)0.9330.9150.915

Cardiotoxicity: QT Prolongation

Description: Drug-induced QT prolongation is defined has having an interval prolongation greater than normal range due to direct pharmacological action to the myocardial cell. Attention has been placed on this condition due to its ability to induce torsades de pointes (TdP), which has the potential to result in sudden cardiac death.

Classification = 1 –> Molecule Indicates Active QT Prolongation

Classification = 0 –> Molecule Demonstrates Inactive QT Prolongation

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9830.6950.734
Specificity1.0000.7410.786
Sensitivity0.9660.6480.682
Area Under the Curve (AUC)0.9990.7560.727

Mutagenicity – TA98 and TA100

Description: A medication can be described as mutagenic if it initiates any of the following three actions within an organism’s DNA: Incorporation of base analogs, specific mis-pairing of bases, or DNA base damage. The Ames Test is the most commonly used method to determine the potential mutagenic effect of a drug. This test typically utilizes 4 test strains of Salmonella typhimurium (TA98, TA100, TA1535, and TA1537) and 1 strain of Escherichia coli (WP2 uvrA) for each assay. The test is conducted in compliance with the Organization for Economic Cooperation and Development (OECD) 471 Guideline for Testing of Chemicals: Bacterial Reverse Mutation Test.

Classification 1 –> Molecule Does Cause Mutagenic Effects

Classification 0 –> Molecule Does Not Cause Mutagenic Effects

Status: Both Models Trained and Tested

Statistics:

TA98 – PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9100.7410.780
Specificity0.8620.7150.727
Sensitivity0.9570.7660.833
Area Under the Curve (AUC)0.9790.7910.851
TA100 – PLS – 3D – SDAR MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9050.6820.710
Specificity0.8500.6220.711
Sensitivity0.9590.7420.708
Area Under the Curve (AUC)0.9770.7640.721

Nephrotoxicity – Interstitial Nephritis (IN)

Description: IN is a kidney condition characterized by swelling between the kidney tubules and is generally the result of an allergic reaction. More than 100 different medications have the capacity to trigger IN, with many of these medications being commonly prescribed classes, such as antibiotics, NSAIDs, and protein pump inhibitors. Since diagnosis of this condition can be a complicated process, having the ability to correctly predict IN activity in a molecule can prove to be exceedingly useful in the drug development process.

Classification = 1 –> Molecule Induces IN

Classification = 0 –> Molecule Does Not Induce IN

Status: Model Trained and Tested

Statistics:

ANN – ECFP – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9850.8430.871
Specificity0.9660.6550.923
Sensitivity0.9880.8750.861
Area Under the Curve (AUC)0.9880.8440.932

Receptor Modeling

Receptor Modeling Description: Treating specific CNS conditions hinge on a drug’s ability to appropriately attach to the receptor associated with the condition. Below is a selection of the receptors Spektron Systems has investigated, with respect to modeling building.

5-HT2A-Agonist vs Antagonist

Description: The mammalian 5-HT2A receptor is a subtype of the 5-HT2 receptor, which belongs to the serotonin receptor family and is a G protein-coupled receptor (GPCR) (Cook et al, 1994). This receptor was first noted for its importance as a target of serotonergic psychedelic drugs such as LSD. The receptor came back to prominence due to its partial mediating action of many antipsychotic drugs, especially the atypically classified.

Classification = 1 –> Full Agonist

Classification = 0 –> Antagonist

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy1.0000.7140.759
Specificity1.0000.7390.793
Sensitivity1.0000.6880.725
Area Under the Curve1.0000.7920.855

5-HT2A-Ketanserin-Binding Affinity

Description: Sources for this data indicate the collection of data was for human receptors utilizing ketanserin as the hot ligand.

Classification = 1 –> Binding Affinity < 200pki

Classification = 0 –> Binding Affinity >200pki

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9420.7820.850
Specificity0.9780.8000.957
Sensitivity0.8940.7580.706
Area Under the Curve (AUC)0.9800.8020.818

5-HT2A – Functionality

Description: This collection of data and model building focuses more on the general classification of functionality regarding molecules. This model set out to predict whether a molecule acts as an agonist or antagonist for the 5-HT2A receptor.

Classification = 1 –> Agonist

Classification = 0 –> Antagonist

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9740.7820.800
Specificity0.9570.7830.750
Sensitivity1.0000.7810.875
Area Under the Curve (AUC)1.0000.8410.969

5-HT7 – LSD – Binding Affinity

Description: The 5-HT7 receptor plays a role in smooth muscle relaxation within the vasculature and in the gastrointestinal tract. This receptor is also involved in thermoregulation, circadian rhythm, learning and memory, and sleep. There is also speculation that this receptor may be involved in mood regulation, suggesting that it may be a useful target in the treatment of depression (Hedlund and Sutcliffe, 2004) (Naumenko et al, 2014). For this model, sources chose LSD as the hot ligand for measuring binding affinity.

Classification = 1 –> Binding Affinity < 1000

Classification= 0 –> Binding Affinity >1000

Status: Model Trained and Tested

Statistics:

XGBoost – ECFP – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9280.8700.844
Specificity0.9110.8750.865
Sensitivity1.0000.8460.750
Area Under the Curve (AUC)0.9620.9180.885

Dopamine (D2) – Methylspiperone – Binding Affinity

Description: GPC dopamine receptors such as D1, D2, D3, D4, and D5 mediate all of the physiological functions of the catecholaminergic neurotransmitter dopamine, ranging from voluntary movement and reward to hormonal regulation and hypertension. Pharmacological agents targeting dopaminergic neurotransmission have been clinically used in the management of several neurological and psychiatric disorders, including Parkinson’s disease, schizophrenia, bipolar disorder, Huntington’s disease, attention deficit hyperactivity disorder (ADHD), and Tourette’s syndrome. Almost all of the clinically effect antipsychotics block D2 dopamine receptors. Atypical antipsychotics exert the same therapeutic effects of the traditional antipsychotics without producing the undesirable side effects of the older, typical antipsychotics. Atypical antipsychotics tend to illustrate a lower affinity for D2 receptors, yet readily bind with D3 and D4 dopamine receptors. Both D2 and D3 receptors fall under the class of D2-like receptors.

This dataset includes binding affinities collected using Methylspiperone as the hot ligand.

Classification = 1 –> Binding Affinity <500pki

Classification = 0 –> Binding Affinity >500pki

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy1.0000.8100.909
Specificity1.0000.8421.000
Sensitivity1.0000.7830.833
Area Under the Curve (AUC)1.0000.8760.867

Dopamine (D2L) – Binding Affinity

Description: This model focuses on the isomer D2L. For this model, sources chose [3]-Spiperone as the hot ligand for measuring binding affinity.

Classification = 1 –> Active D2L Binding

Classification = 0 –> Inactive D2L Binding

Status: Model Trained and Pending Approval

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9720.7220.750
Specificity1.0000.7690.833
Sensitivity0.9660.7120.731
Area Under the Curve (AUC)1.0000.8310.776

Dopamine (D2) – Functionality

Description: This collection of data and model building focuses on the general classification of functionality regarding molecules. This model aims to predict whether a molecule acts as an agonist or antagonist for the D2 receptor.

Classification = 1 –> Agonist

Classification = 0 –> Antagonist

Status: Model Trained and Pending Approval

Statistics:

ANN – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy1.0000.8130.824
Specificity1.0000.7410.708
Sensitivity1.0000.8850.939
Area Under the Curve (AUC)1.0000.8660.869

Dopamine (D3) – [3H] – Spiperone – Binding Affinity

Description: Sources tested binding affinities for Dopamine (D3) using the hot ligand [3H] – Spiperone.

Classification = 1 –> Binding Affinity <100pki

Classification = 0 –> Binding Affinity >100pki

Status: Model Trained and Pending Approval

Statistics:

PLS – ECFP – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy1.0000.8960.837
Specificity1.0000.8490.923
Sensitivity1.0000.9170.800
Area Under the Curve (AUC)1.0000.9310.926

Opioid Reversal – Delta Receptor – Binding Affinity

Description: Genetic approaches have corroborated the importance of delta opioid receptors in chronic pain. Pharmacological and genetic data highlight delta opioid agonists as a potential alternative to mu analgesics for treatment of chronic pain (Pradhan et al, 2011). This model explores the ability to discriminate between molecules that are likely to be binders or non-binders to the delta opioid receptor.

Classification = 1 –> Strong Binding Activity <100pki

Classification = 0 –> Weak-No Binding Activity >100pki

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.8910.8490.848
Specificity0.8620.8190.807
Sensitivity0.9200.8780.888
Area Under the Curve (AUC)0.9580.9170.921

Opioid Reversal – Delta Receptor – Functionality

Description: This collection of data and model building focuses on the general classification of functionality regarding molecules. This model aims to predict whether a molecule acts as an agonist or antagonist for the delta opioid receptor.

Classification = 1 –> Antagonist Functionality

Classification = 0 –> Agonist Functionality

Status: Model Trained and Pending Approval

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.9430.7390.917
Specificity0.8860.7711.000
Sensitivity1.0000.7060.833
Area Under the Curve (AUC)1.0000.7550.931

Opioid Reversal – Kappa Receptor – Binding Affinity

Description: In addiction models, the activity of the kappa receptor is potentiated by stressors and plays a strong role in drug-seeking and relapse (Lalanne et al, 2014). This model explores the ability to discriminate between molecules that are likely to be binders or non-binders to the kappa opioid receptor.

Classification = 1 –> Strong Binding Activity </=100pki

Classification = 0 –> Weak-No Binding Activity >100pki

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.8760.8300.804
Specificity0.9180.8770.857
Sensitivity0.8330.7820.750
Area Under the Curve (AUC)0.9400.8860.866

Opioid Reversal – Kappa Receptor – Functionality

Description: This collection of data and model building focuses on the general classification of functionality regarding molecules. This model aims to predict whether a molecule acts as an agonist or antagonist for the kappa opioid receptor.

Classification = 1 –> Antagonist Functionality

Classification = 0 –> Agonist Functionality

Status: Model Trained and Pending Approval

Statistics:

PLS – 3D -SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy1.0000.7590.818
Specificity1.0000.7930.867
Sensitivity1.0000.7250.769
Area Under the Curve (AUC)1.0000.7830.913

Opioid Reversal – Mu Receptor – Binding Affinity

Description: Of the three opioid receptor families, the mu opioid receptor subtypes have been most extensively studied due to their role in mediating the actions of morphine and other clinically relevant pain-relieving agents, in addition to drugs that are more likely to be abused (Feng et al, 2013).

Classification = 1 –> Strong Binding Activity <100pki

Classification = 0 –> Weak-No Binding Activity >100pki

Status: Model Trained and Tested

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy0.8620.8060.822
Specificity0.8580.7990.819
Sensitivity0.8650.8120.824
Area Under the Curve (AUC)0.9370.8850.890

Opioid Reversal – Mu Receptor – Functionality

Description: This collection of data and model building focuses on the general classification of functionality regarding molecules. This model aims to predict whether a molecule acts as an agonist or antagonist for the mu opioid receptor.

Classification = 1 –> Antagonist Functionality

Classification = 0 –> Agonist Functionality

Status: Model Training and Pending Approval

Statistics:

PLS – 3D – SDAR – MetricsTraining ScoreValidation ScoreTest Score
Balanced Accuracy1.0000.8060.814
Specificity1.0000.7990.733
Sensitivity1.0000.8120.895
Area Under the Curve (AUC)1.0000.8850.912