Welcome to SMAG

Smart Molecule AI-Driven Generation

Salient Points Benefits
About SMAG AI-Driven Molecular Generation Technology
We are redefining the future of drug discovery with our AI-driven molecular generation technology

Our mission is to empower researchers and pharmaceutical innovators with precise, efficient, and scalable tools to accelerate the identification of novel compounds and therapeutic solutions. By combining patented algorithms with advanced artificial intelligence, we unlock unprecedented opportunities for breakthroughs in drug design and development.

At SMAG, we offer state-of-the-art models

This includes Data Generation, QSAR, Toxicity Prediction, Retrosynthesis, and Drug Repurposing, designed to revolutionize molecular research across diverse industries. From utilizing Graph Transformers for designing molecules with superior novelty and uniqueness to deploying cutting-edge predictive models for toxicity and retrosynthesis, our technology consistently outperforms the competition

Topia Life Sciences developed a world-class AI platform that delivers unmatched accuracy and efficiency in molecule generation

Through proprietary datasets and advanced processing, SMAG ensures results that lead to impactful discoveries.

"Through the advancement of our platforms, we will build a fully integrated data economy. At Topia Life Sciences, we are passionate about improving health outcomes for people around the world" - Mr. Kamlesh Patel, Founder

Our Services

The purpose of the Molecule Generation model is to revolutionize the process of molecular design and discovery using advanced deep learning techniques.

Data Generation

Topia's model uses generative modeling to accelerate the creation of small molecules for drug discovery and materials science, identifying novel compounds and optimizing drug candidates for various industry applications.

Key considerations:

  • Data cleaning and Filtering
  • Data Standardization and Normalization
  • Data Fragmentation
  • Data Encoding

Molecule Generation Model

  • Utilizes Graph Transformer for molecule generation
  • Achieves 100% validity in generated molecules, compared to 87-92% by State-of-the-Art (SOTA) models
  • Achieves 98% uniqueness, 99% accuracy, and 97% novelty in generated molecules, outperforming SOTA models in all these metrics.
  • Incorporates parallel processing for enhanced performance

QSAR

Topia’s Quantitative Structure-Activity Relationship (QSAR) model stands as a pillar of innovation.Offering a meticulous understanding of how molecular features correlate with biological activity. Our computational tools harness the power of advanced algorithms to decode the intricate relationship between chemical structure and functional behaviour.

Key considerations:

  • Molecular Descriptors
  • Model Building
  • Validation: Calculation of Molecular Fingerprints, Fingerprint Filtering, Data Splitting, Machine Learning Regression, Model Evaluation

QSAR Model

  • Predictive Performance
  • Flexibility and Customization
  • Scalability and Efficiency
  • Interpretability and Transparency

Toxicity:

Plays a crucial role in ensuring the safety and efficacy of various chemicals and compounds across industries. Our Toxicity prediction model is a multifaceted endeavour, encompassing various domains such as mutagenicity, carcinogenicity, nephrotoxicity, cardiotoxicity, and hepatotoxicity.

Key considerations:

  • Data Collection and Preprocessing
  • Model Architecture Selection
  • Training and Validation
  • Computational Efficiency and Scalability
  • External Validation and Benchmarking

Toxicity Model

  • Our toxicity prediction models consistently outperform existing models.
  • Consistently achieve higher true positive and negative rates while minimizing false positives and negatives.
  • The ROC curve (AUC) for our in-house models is consistently higher than existing models.
  • Our toxicity prediction models cover a range of toxicological endpoints, including mutagenicity, carcinogenicity, nephrotoxicity, cardiotoxicity, and hepatotoxicity.

Retrosynthesis:

Predicts the synthetic pathway of small organic molecules, empowering chemists to devise efficient synthesis strategies.
The retrosynthesis model developed by Topia Life Sciences incorporates unique methodologies rooted in AI and Chemical Informatics principles. The model employs a combination of two models wherein the first model is based on Multi-Layer Perceptron that predicts the single step reactions and the second model is based on A* search algorithm wherein this is used to find the shortest distance between two nodes of a graph.

Key considerations:

  • Data Cleaning and Standardization
  • Reaction Centre Identification
  • Validation and Quality Control

Retrosynthesis Model

  • Both run in parallel to search for the commercial availability of the reagents required
  • Predict higher Accuracy for Drug Structures
  • Predict higher Accuracy for Modified Structures
  • Route Similarity with State-of-the-Art Models
  • Feasibility of Predicted Routes

Drug Repurpose:

Expedite the drug discovery process by repurposing existing drugs for new therapeutic uses. Our AI model encompasses the development and validation of a Drug Repurposing pipeline, integrating Knowledge Graph embeddings with Machine Learning algorithms.

Key considerations:

  • Knowledge Graph Embeddings (KGE)
  • Machine Learning (ML) Algorithms
  • Novel Approaches and Advancements
  • Data Preprocessing Techniques

Repurpose Model

  • Our AI model utilizes a comprehensive biomedical knowledge graph with diverse entities (drugs, diseases, proteins, genes, etc.) and relations, providing a rich source of information, in the form of a fact-based graph which can be used as a reference in drug repurposing.
  • Our model can learn complex patterns and relationships, enabling accurate prediction of novel drug-disease associations.
  • Overcome some of the issues of Knowledge graphs and Graphs such as sparse and dense subgraphs
  • Proprietary Datasets
  • With a proven track record of successful projects.

Salient Points

SMAG can generate novel molecules for any target, instead of screening the chemical space

Shortlisting the hit molecules with data-centric method

Predicting the toxicity based on the computational model built on commercial data.

Score Predicting the ADME properties and Toxicity based on the AI models

Enhanced Efficiency

Increased Accuracy

Cost-Effective Solutions

Flexibility and Adaptability

Predictive Modelling

Our Presence:

United Kingdom

128 City Road, London, United Kingdom, EC1V 2NX

United States

1710 East Franklin Street #1137, Chapel Hill, NC 27514

India

Meldi Estate, 22B, Railway Crossing Rd, near Gota, Gota, Ahmedabad, Gujarat 382481, India

India

Siddarth Annexure 3, National Highway 8, beside Hotel Plazzo, Vemali, Vadodara, Gujarat 391740, India

Romania

Str. Panselelor, nr. 27-28,Brasov - 500419, Romania