Machine Learning
Machine Learning
Generative models are increasingly used to predict drug activity, allowing researchers to design new molecules and optimize existing ones by simulating their interactions with biological targets. These models excel in scenarios where large amounts of data are available, providing insights into how a molecule’s structure might influence its behavior as a drug. Here’s a breakdown of the main generative models used for drug activity prediction:
There are main types of machine learning :
- Generative Adversarial Networks (GANs).
- Variational Autoencoders (VAEs).
- Recurrent Neural Networks (RNNs).
- Transformer Models.
- Graph-Based Generative Models.
Generative Adversarial Networks (GANs)
GANs consist of two neural networks—a generator and a discriminator—competing to improve realistic data generation. In drug discovery, GANs can generate novel molecular structures that could potentially interact with a biological target. The generator network creates new molecular structures, while the discriminator evaluates them to ensure they resemble known active compounds.
Applications:
- Creating new drug-like molecules with desired chemical properties.
- Exploring the chemical space more efficiently by generating diverse molecules for screening.
Example: GANs have generated molecules optimized for specific properties, such as bioactivity or solubility, based on known training data.
Variational Autoencoders (VAEs)
VAEs are another popular generative model that learns latent representations of molecules. They take molecular data (e.g., SMILES notation of a compound) and encode it into a lower-dimensional latent space, where variations can be made to create novel molecules. The decoder then reconstructs these variations back into valid chemical structures.
Applications:
- Designing molecules with improved pharmacological properties.
- Modifying existing compounds to optimize drug activity by exploring the latent space.
Example: VAEs have been applied to create molecules with predicted high-binding affinity to specific protein targets, improving lead compound generation.
Recurrent Neural Networks (RNNs)
RNNs, particularly Long Short-Term Memory (LSTM) networks, are generative models well-suited for sequence data like SMILES strings. RNNs learn the sequence patterns of molecular structures and generate new molecules that follow the learned patterns.
Applications:
- Generation of drug-like compounds by learning the grammar and syntax of chemical structures.
- Prediction of drug activity by generating molecules with similar sequences to known active compounds.
Example: RNN-based models have been used to generate molecules that satisfy specific activity constraints, making them useful for creating novel drug candidates.
Transformer Models
Transformers, a newer type of deep learning architecture, have shown great promise in generating molecular structures. These models, initially developed for natural language processing (NLP), can process SMILES sequences to predict properties or generate novel compounds with desired biological activities.
Applications:
- Generating chemically valid drug-like molecules.
- Optimizing the structure of drug candidates by focusing on activity-related features.
Example: Transformer-based models have been used to predict chemical reactions and propose modifications to drug molecules to enhance their biological activity.
Graph-Based Generative Models
Since molecules can be represented as graphs (atoms as nodes and bonds as edges), generative models like Graph Neural Networks (GNNs) have been applied to drug discovery. These models generate new molecular graphs with desired properties such as activity, solubility, or toxicity.
Applications:
- Predicting drug activity by generating molecular graphs that interact with biological targets.
- De novo drug design by exploring molecular structures that meet specific bioactivity criteria.
Example: Graph-based models are used to predict the biological activity of new compounds based on the structure-activity relationship (SAR) learned from existing molecules.
What should you know about our AI models for
predicting molecular bioactivity?
- The models achieve an accuracy of 85-95%.
- The compound's SMILES code is utilized for data entry to forecast activity.
- Designed to predict chemical bioactivity, the models focus on nanomolar potency. If the chemical properties of the test molecule do not align with the model's feature scheme or if the model predicts micromolar potency, the compound will be considered inactive.
Our AI Models
Our AI Models
BACE2 Inhibitors
BACE2 (Beta-Site APP-Cleaving Enzyme 2) inhibitors are gaining attention for their therapeutic potential, particularly in conditions like type 2 diabetes and Alzheimer’s disease. Unlike BACE1, which is primarily implicated in the production of amyloid-beta peptides in Alzheimer’s, BACE2 plays roles in insulin regulation and other pathways.
Notable BACE2 Inhibitors
- Small Molecule Inhibitors:
- Early inhibitors targeting BACE enzymes often lacked specificity but provided a basis for optimizing BACE2-selective compounds.
- Compounds based on azepine, hydroxyethylamine, or other scaffolds have shown promise in selectively targeting BACE2.
- Peptidomimetic Inhibitors:
- These are designed to mimic the natural substrates of BACE2, offering higher specificity and potency.
Challenges in BACE2 Inhibitor Discovery
Phosphodiesterase 5 (PDE5) inhibitors
The discovery of Phosphodiesterase 5 (PDE5) inhibitors marked a major milestone in medicine, particularly in the treatment of erectile dysfunction (ED) and pulmonary arterial hypertension (PAH).
In 1998, the FDA approved sildenafil (Viagra) for Pfizer, making it the first PDE5 inhibitor for ED treatment. Tadalafil (Cialis) (Approved in 2003): Known for its long duration of action, lasting up to 36 hours. Vardenafil (Levitra) (Approved in 2003): Offers rapid onset and high selectivity for PDE5. Avanafil (Stendra) (Approved in 2012): Features a faster onset and shorter duration, providing more flexibility.
Beyond ED, PDE5 inhibitors were discovered to have therapeutic effects in other conditions:
- Pulmonary Arterial Hypertension (PAH): Sildenafil (Revatio) and tadalafil were approved for PAH as they reduce pulmonary vascular resistance.
- Other Investigations: PDE5 inhibitors are being studied for their potential in treating conditions like Raynaud's phenomenon, heart failure, and even some cancers.
BTK Inhibitors
The discovery of Bruton’s Tyrosine Kinase (BTK) inhibitors represents a significant breakthrough in targeted therapies for diseases like B-cell malignancies, autoimmune disorders, and other immune-mediated conditions. BTK is a critical enzyme in the B-cell receptor (BCR) signaling pathway. It regulates B-cell maturation, activation, and survival, making it a key target in diseases involving dysregulated B cells (e.g., chronic lymphocytic leukemia, mantle cell lymphoma). BTK was discovered in 1993 as the gene mutated in X-linked agammaglobulinemia (XLA), a condition marked by an absence of mature B cells.
Ibrutinib (PCI-32765): The first BTK inhibitor approved by the FDA (2013). Discovered by Pharmacyclics, it irreversibly binds to the cysteine residue (C481) in BTK’s ATP-binding site. Ibrutinib is highly effective in treating chronic lymphocytic leukemia (CLL), mantle cell lymphoma (MCL), and Waldenström’s macroglobulinemia (WM). Acalabrutinib: A more selective, irreversible BTK inhibitor with reduced off-target effects. Approved for CLL and MCL. Zanubrutinib: Aimed at further improving specificity and reducing side effects. Pirtobrutinib (LOXO-305): A reversible (non-covalent) BTK inhibitor designed to overcome resistance mutations like C481S.
CDK9 Inhibitors
The discovery of CDK9 (Cyclin-Dependent Kinase 9) inhibitors has garnered significant attention due to their potential to regulate transcription and treat cancers, inflammatory diseases, and viral infections. CDK9 is a key component of the transcriptional machinery and offers a unique target for therapeutic intervention. Targeting CDK9 can selectively inhibit the transcription of genes with short half-lives, particularly those critical for cancer cell survival and proliferation. CDK9 regulates the expression of short-lived anti-apoptotic proteins like MCL-1 and MYC, making it a promising target for cancers dependent on these proteins, it influences inflammatory gene transcription, and Many viruses, including HIV, hijack CDK9 to enhance viral transcription.
Flavopiridol (alvocidib): The first-generation pan-CDK inhibitor with activity against CDK9. Though nonspecific, flavopiridol demonstrated potent transcriptional repression and entered clinical trials for cancers like acute myeloid leukemia (AML). Lacking the selectivity for CDK9 over other CDKs (e.g., CDK2, CDK4/6), compensatory pathways that bypass CDK9, and the off-target toxicities, particularly gastrointestinal and hematological effects, are major challenges of the CDK9 inhibitors that need to be overcome in their discovery cycles.
BRD4 Inhibitors
The discovery of Bromodomain-Containing Protein 4 (BRD4) inhibitors has revolutionized targeted therapy approaches, especially in cancer and inflammation. BRD4, a member of the bromodomain and extra-terminal (BET) family, is a key regulator of transcriptional programs, making it an attractive drug target. BRD4 binds to acetylated lysine residues on histones via its bromodomains (BD1 and BD2), facilitating chromatin remodeling. It recruits transcriptional machinery, including P-TEFb, to regulate transcriptional elongation of key oncogenes like MYC. Targeting BRD4 could disrupt oncogenic transcriptional programs and immune-inflammatory signaling.
JQ1: A seminal discovery by James E. Bradner and colleagues, is a potent, reversible BRD4 inhibitor that competitively binds to the acetyl-lysine recognition pocket of the bromodomain. It showed strong preclinical efficacy in MYC-driven cancers, such as NUT midline carcinoma (NMC) and hematologic malignancies. BRD4 inhibitors are effective in multiple myeloma, acute myeloid leukemia (AML), diffuse large B-cell lymphoma (DLBCL), and MYC-driven cancers, such as NUT midline carcinoma and triple-negative breast cancer. BRD4 inhibitors are being explored in rheumatoid arthritis, inflammatory bowel disease, sepsis, and Anti-fibrotic potential in cardiac and pulmonary fibrosis. Resistance mechanisms and dose-limiting toxicities, including thrombocytopenia and gastrointestinal effects, necessitate the development of selective or context-specific inhibitors.
Chemokine receptor (CC) antagonists
The discovery of chemokine receptor (CC) antagonists represents a critical step in targeting immune responses, inflammation, and cancer. CC chemokine receptors are a subset of the G-protein-coupled receptor (GPCR) family that mediate the activity of CC chemokines, which regulate immune cell trafficking. Dysregulation of CC chemokine signaling has been implicated in several pathological conditions, including autoimmune diseases, cancer, and viral infections. CC chemokine receptors (e.g., CCR1, CCR2, CCR3, CCR4) bind CC chemokines, which contain adjacent cysteine residues near their amino terminus. Abnormal signaling promotes leukocyte infiltration in diseases like rheumatoid arthritis and multiple sclerosis. Chemokines like CCL2 (via CCR2) recruit tumor-associated macrophages, fostering tumor progression and metastasis. Certain CC receptors, like CCR5, serve as entry points for viruses such as HIV. Blocking CC chemokine receptor signaling can modulate immune cell migration and reduce inflammation or tumor-supportive microenvironments.
Maraviroc is the first FDA-approved CCR5 antagonist for HIV treatment. PF-04136309 is a clinical-stage CCR2 antagonist tested for pancreatic cancer and rheumatoid arthritis. CCR3 is involved in eosinophil recruitment in allergic diseases, and GW766994 demonstrated efficacy in preclinical asthma models.
Toll-like receptor (TLR7,8) agonists
The discovery of Toll-like receptor (TLR) 7 and TLR8 agonists has opened new avenues for immunotherapy, vaccine adjuvants, and antiviral treatments. These innate immune receptors play critical roles in recognizing single-stranded RNA (ssRNA) and initiating immune responses. Their agonists have shown promise in oncology, infectious diseases, and autoimmune conditions. TLR7 and TLR8 are endosomal receptors in the innate immune system. They recognize viral ssRNA or synthetic RNA analogs and activate signaling pathways (e.g., NF-κB, IRF7) leading to cytokine production and immune activation. Activation of TLR7/8 can enhance antigen presentation, promote cytotoxic T-cell responses, and stimulate anti-tumor immunity. TLR7/8 agonists mimic viral infections to boost immune responses against pathogens.
Imiquimod (TLR7 agonist) is FDA-approved for treating actinic keratosis, superficial basal cell carcinoma, and genital warts. Resiquimod (TLR7/8 agonist) is used in preclinical and clinical studies for cancer and infectious diseases. GS-9620 (vesatolimod) is a selective TLR7 agonist with antiviral activity against HIV and HBV. GS-9688 is a TLR8 agonist targeting chronic HBV.
Overactivation of TLR7/8 can lead to cytokine storms, flu-like symptoms, and systemic inflammation, achieving receptor selectivity to minimize off-target effects, and Developing formulations for targeted delivery (e.g., nanoparticles, liposomes).
Histone Deacetylase 6 (HDAC6) inhibitors
The discovery of Histone Deacetylase 6 (HDAC6) inhibitors marks a significant advancement in targeted therapies for cancer, neurodegenerative diseases, and autoimmune conditions. HDAC6 is a unique member of the histone deacetylase family, with noncanonical roles that make it a compelling drug target. Unlike nuclear HDACs, HDAC6 is predominantly cytoplasmic and deacetylates non-histone proteins such as α-tubulin (microtubule dynamics), Hsp90 (protein stability and function), Cortactin (cell motility), and plays roles in protein degradation via the aggresome-autophagy pathway.
ACY-1215 (Ricolinostat) is the first-in-class selective HDAC6 inhibitor to enter clinical trials and demonstrated synergy with proteasome inhibitors like bortezomib in multiple myeloma. ACY-241 (Citarinostat) is A next-generation HDAC6 inhibitor with improved pharmacokinetics and broader clinical applications. Achieving HDAC6 specificity while avoiding off-target effects on other HDAC isoforms, cytopenia, gastrointestinal symptoms, and neurotoxicity observed with pan-HDAC inhibitors is less pronounced but still a concern and adaptive changes in cancer cells may reduce the efficacy, are major challenges of the HDAC6 inhibitors that need to be overcome in their discovery cycles through innovative molecular designs.
BACE2 (Beta-Site APP-Cleaving Enzyme 2) inhibitors are gaining attention for their therapeutic potential, particularly in conditions like type 2 diabetes and Alzheimer’s disease. Unlike BACE1, which is primarily implicated in the production of amyloid-beta peptides in Alzheimer’s, BACE2 plays roles in insulin regulation and other pathways.
Notable BACE2 Inhibitors
- Small Molecule Inhibitors:
- Early inhibitors targeting BACE enzymes often lacked specificity but provided a basis for optimizing BACE2-selective compounds.
- Compounds based on azepine, hydroxyethylamine, or other scaffolds have shown promise in selectively targeting BACE2.
- Peptidomimetic Inhibitors:
- These are designed to mimic the natural substrates of BACE2, offering higher specificity and potency.
Challenges in BACE2 Inhibitor Discovery
The discovery of Phosphodiesterase 5 (PDE5) inhibitors marked a major milestone in medicine, particularly in the treatment of erectile dysfunction (ED) and pulmonary arterial hypertension (PAH).
In 1998, the FDA approved sildenafil (Viagra) for Pfizer, making it the first PDE5 inhibitor for ED treatment. Tadalafil (Cialis) (Approved in 2003): Known for its long duration of action, lasting up to 36 hours. Vardenafil (Levitra) (Approved in 2003): Offers rapid onset and high selectivity for PDE5. Avanafil (Stendra) (Approved in 2012): Features a faster onset and shorter duration, providing more flexibility.
Beyond ED, PDE5 inhibitors were discovered to have therapeutic effects in other conditions:
- Pulmonary Arterial Hypertension (PAH): Sildenafil (Revatio) and tadalafil were approved for PAH as they reduce pulmonary vascular resistance.
- Other Investigations: PDE5 inhibitors are being studied for their potential in treating conditions like Raynaud’s phenomenon, heart failure, and even some cancers.
he discovery of Bruton’s Tyrosine Kinase (BTK) inhibitors represents a significant breakthrough in targeted therapies for diseases like B-cell malignancies, autoimmune disorders, and other immune-mediated conditions. BTK is a critical enzyme in the B-cell receptor (BCR) signaling pathway. It regulates B-cell maturation, activation, and survival, making it a key target in diseases involving dysregulated B cells (e.g., chronic lymphocytic leukemia, mantle cell lymphoma). BTK was discovered in 1993 as the gene mutated in X-linked agammaglobulinemia (XLA), a condition marked by an absence of mature B cells.
Ibrutinib (PCI-32765): The first BTK inhibitor approved by the FDA (2013). Discovered by Pharmacyclics, it irreversibly binds to the cysteine residue (C481) in BTK’s ATP-binding site. Ibrutinib is highly effective in treating chronic lymphocytic leukemia (CLL), mantle cell lymphoma (MCL), and Waldenström’s macroglobulinemia (WM). Acalabrutinib: A more selective, irreversible BTK inhibitor with reduced off-target effects. Approved for CLL and MCL. Zanubrutinib: Aimed at further improving specificity and reducing side effects. Pirtobrutinib (LOXO-305): A reversible (non-covalent) BTK inhibitor designed to overcome resistance mutations like C481S.
The discovery of CDK9 (Cyclin-Dependent Kinase 9) inhibitors has garnered significant attention due to their potential to regulate transcription and treat cancers, inflammatory diseases, and viral infections. CDK9 is a key component of the transcriptional machinery and offers a unique target for therapeutic intervention. Targeting CDK9 can selectively inhibit the transcription of genes with short half-lives, particularly those critical for cancer cell survival and proliferation. CDK9 regulates the expression of short-lived anti-apoptotic proteins like MCL-1 and MYC, making it a promising target for cancers dependent on these proteins, it influences inflammatory gene transcription, and Many viruses, including HIV, hijack CDK9 to enhance viral transcription.
Flavopiridol (alvocidib): The first-generation pan-CDK inhibitor with activity against CDK9. Though nonspecific, flavopiridol demonstrated potent transcriptional repression and entered clinical trials for cancers like acute myeloid leukemia (AML). Lacking the selectivity for CDK9 over other CDKs (e.g., CDK2, CDK4/6), compensatory pathways that bypass CDK9, and the off-target toxicities, particularly gastrointestinal and hematological effects, are major challenges of the CDK9 inhibitors that need to be overcome in their discovery cycles.
BACE2 (Beta-Site APP-Cleaving Enzyme 2) inhibitors are gaining attention for their therapeutic potential, particularly in conditions like type 2 diabetes and Alzheimer’s disease. Unlike BACE1, which is primarily implicated in the production of amyloid-beta peptides in Alzheimer’s, BACE2 plays roles in insulin regulation and other pathways.
Notable BACE2 Inhibitors
- Small Molecule Inhibitors:
- Early inhibitors targeting BACE enzymes often lacked specificity but provided a basis for optimizing BACE2-selective compounds.
- Compounds based on azepine, hydroxyethylamine, or other scaffolds have shown promise in selectively targeting BACE2.
- Peptidomimetic Inhibitors:
- These are designed to mimic the natural substrates of BACE2, offering higher specificity and potency.
Challenges in BACE2 Inhibitor Discovery
The discovery of chemokine receptor (CC) antagonists represents a critical step in targeting immune responses, inflammation, and cancer. CC chemokine receptors are a subset of the G-protein-coupled receptor (GPCR) family that mediate the activity of CC chemokines, which regulate immune cell trafficking. Dysregulation of CC chemokine signaling has been implicated in several pathological conditions, including autoimmune diseases, cancer, and viral infections. CC chemokine receptors (e.g., CCR1, CCR2, CCR3, CCR4) bind CC chemokines, which contain adjacent cysteine residues near their amino terminus. Abnormal signaling promotes leukocyte infiltration in diseases like rheumatoid arthritis and multiple sclerosis. Chemokines like CCL2 (via CCR2) recruit tumor-associated macrophages, fostering tumor progression and metastasis. Certain CC receptors, like CCR5, serve as entry points for viruses such as HIV. Blocking CC chemokine receptor signaling can modulate immune cell migration and reduce inflammation or tumor-supportive microenvironments.
Maraviroc is the first FDA-approved CCR5 antagonist for HIV treatment. PF-04136309 is a clinical-stage CCR2 antagonist tested for pancreatic cancer and rheumatoid arthritis. CCR3 is involved in eosinophil recruitment in allergic diseases, and GW766994 demonstrated efficacy in preclinical asthma models.
The discovery of Toll-like receptor (TLR) 7 and TLR8 agonists has opened new avenues for immunotherapy, vaccine adjuvants, and antiviral treatments. These innate immune receptors play critical roles in recognizing single-stranded RNA (ssRNA) and initiating immune responses. Their agonists have shown promise in oncology, infectious diseases, and autoimmune conditions. TLR7 and TLR8 are endosomal receptors in the innate immune system. They recognize viral ssRNA or synthetic RNA analogs and activate signaling pathways (e.g., NF-κB, IRF7) leading to cytokine production and immune activation. Activation of TLR7/8 can enhance antigen presentation, promote cytotoxic T-cell responses, and stimulate anti-tumor immunity. TLR7/8 agonists mimic viral infections to boost immune responses against pathogens.
Imiquimod (TLR7 agonist) is FDA-approved for treating actinic keratosis, superficial basal cell carcinoma, and genital warts. Resiquimod (TLR7/8 agonist) is used in preclinical and clinical studies for cancer and infectious diseases. GS-9620 (vesatolimod) is a selective TLR7 agonist with antiviral activity against HIV and HBV. GS-9688 is a TLR8 agonist targeting chronic HBV.
Overactivation of TLR7/8 can lead to cytokine storms, flu-like symptoms, and systemic inflammation, achieving receptor selectivity to minimize off-target effects, and Developing formulations for targeted delivery (e.g., nanoparticles, liposomes).
The discovery of Histone Deacetylase 6 (HDAC6) inhibitors marks a significant advancement in targeted therapies for cancer, neurodegenerative diseases, and autoimmune conditions. HDAC6 is a unique member of the histone deacetylase family, with noncanonical roles that make it a compelling drug target. Unlike nuclear HDACs, HDAC6 is predominantly cytoplasmic and deacetylates non-histone proteins such as α-tubulin (microtubule dynamics), Hsp90 (protein stability and function), Cortactin (cell motility), and plays roles in protein degradation via the aggresome-autophagy pathway.
ACY-1215 (Ricolinostat) is the first-in-class selective HDAC6 inhibitor to enter clinical trials and demonstrated synergy with proteasome inhibitors like bortezomib in multiple myeloma. ACY-241 (Citarinostat) is A next-generation HDAC6 inhibitor with improved pharmacokinetics and broader clinical applications. Achieving HDAC6 specificity while avoiding off-target effects on other HDAC isoforms, cytopenia, gastrointestinal symptoms, and neurotoxicity observed with pan-HDAC inhibitors is less pronounced but still a concern and adaptive changes in cancer cells may reduce the efficacy, are major challenges of the HDAC6 inhibitors that need to be overcome in their discovery cycles through innovative molecular designs.