fantasia 2.1.0

Functional ANnoTAtion based on embedding space SImilArity

FANTASIA

Functional ANnoTAtion based on embedding space SImilArity

FANTASIA is an advanced pipeline for the automatic functional annotation of protein sequences using state-of-the-art protein language models. It integrates deep learning embeddings and in-memory similarity searches, retrieving reference vectors from a PostgreSQL database with pgvector, to associate Gene Ontology (GO) terms with proteins.

For full documentation, visit FANTASIA Documentation.

Key Features

• ✅ Available Embedding Models
  Supports protein language models: ProtT5, ProstT5, and ESM2 for sequence representation.

• 🔍 Redundancy Filtering
  Filters out homologous sequences using CD-HIT in the lookup table, allowing controlled redundancy levels through an adjustable threshold, ensuring reliable benchmarking and evaluation.

• 💾 Optimized Data Storage
  Embeddings are stored in HDF5 format for input sequences. The reference table, however, is hosted in a public relational PostgreSQL database using pgvector.

• 🚀 Efficient Similarity Lookup
  Performs high-speed searches using in-memory computations. Reference vectors are retrieved from a PostgreSQL database with pgvector for comparison.

• 🔬 Functional Annotation by Similarity
  Assigns Gene Ontology (GO) terms to proteins based on embedding space similarity, using pre-trained embeddings.

Pipeline Overview (Simplified)

1. Embedding Generation
   Computes protein embeddings using deep learning models (ProtT5, ProstT5, and ESM2).

2. GO Term Lookup
   Performs vector similarity searches using in-memory computations to assign Gene Ontology terms. Reference embeddings are retrieved from a PostgreSQL database with pgvector. Only experimental evidence codes are used for transfer.

Acknowledgments

FANTASIA is the result of a collaborative effort between Ana Rojas’ Lab (CBBIO) (Andalusian Center for Developmental Biology, CSIC) and Rosa Fernández’s Lab (Metazoa Phylogenomics Lab, Institute of Evolutionary Biology, CSIC-UPF). This project demonstrates the synergy between research teams with diverse expertise.

This version of FANTASIA builds upon previous work from:

• Metazoa Phylogenomics Lab's FANTASIA
  The original implementation of FANTASIA for functional annotation.

• bio_embeddings
  A state-of-the-art framework for generating protein sequence embeddings.

• GoPredSim
  A similarity-based approach for Gene Ontology annotation.

• protein-metamorphisms-is
  Serves as the reference biological information system, providing a robust data model and curated datasets for protein structural and functional analysis.

We also extend our gratitude to LifeHUB-CSIC for inspiring this initiative and fostering innovation in computational biology.

Citing FANTASIA

If you use FANTASIA in your research, please cite the following publications:

1. Martínez-Redondo, G. I., Barrios, I., Vázquez-Valls, M., Rojas, A. M., & Fernández, R. (2024).
   Illuminating the functional landscape of the dark proteome across the Animal Tree of Life.
   DOI: 10.1101/2024.02.28.582465

2. Barrios-Núñez, I., Martínez-Redondo, G. I., Medina-Burgos, P., Cases, I., Fernández, R., & Rojas, A. M. (2024).
   Decoding proteome functional information in model organisms using protein language models.
   DOI: 10.1101/2024.02.14.580341

---

👥 Project Team

🔧 Technical Team

• Francisco Miguel Pérez Canales: [e-mail]
  Author of the system’s engineering and technical implementation
• Francisco J. Ruiz Mota: [e-mail]
  Junior developer

🧬 Scientific Team & Original Authors of FANTASIA v1

• Ana M. Rojas: [e-mail]
• Gemma I. Martínez-Redondo: [e-mail]
• Rosa Fernández: [e-mail]

---

FANTASIA

Functional ANnoTAtion based on embedding space SImilArity

FANTASIA is an advanced pipeline for the automatic functional annotation of protein sequences using state-of-the-art protein language models. It integrates deep learning embeddings and in-memory similarity searches, retrieving reference vectors from a PostgreSQL database with pgvector, to associate Gene Ontology (GO) terms with proteins.

For full documentation, visit FANTASIA Documentation.

Key Features

• ✅ Available Embedding Models
  Supports protein language models: ProtT5, ProstT5, and ESM2 for sequence representation.

• 🔍 Redundancy Filtering
  Filters out homologous sequences using CD-HIT in the lookup table, allowing controlled redundancy levels through an adjustable threshold, ensuring reliable benchmarking and evaluation.

• 💾 Optimized Data Storage
  Embeddings are stored in HDF5 format for input sequences. The reference table, however, is hosted in a public relational PostgreSQL database using pgvector.

• 🚀 Efficient Similarity Lookup
  Performs high-speed searches using in-memory computations. Reference vectors are retrieved from a PostgreSQL database with pgvector for comparison.

• 🔬 Functional Annotation by Similarity
  Assigns Gene Ontology (GO) terms to proteins based on embedding space similarity, using pre-trained embeddings.

Pipeline Overview (Simplified)

1. Embedding Generation
   Computes protein embeddings using deep learning models (ProtT5, ProstT5, and ESM2).

2. GO Term Lookup
   Performs vector similarity searches using in-memory computations to assign Gene Ontology terms. Reference embeddings are retrieved from a PostgreSQL database with pgvector. Only experimental evidence codes are used for transfer.

Acknowledgments

FANTASIA is the result of a collaborative effort between Ana Rojas’ Lab (CBBIO) (Andalusian Center for Developmental Biology, CSIC) and Rosa Fernández’s Lab (Metazoa Phylogenomics Lab, Institute of Evolutionary Biology, CSIC-UPF). This project demonstrates the synergy between research teams with diverse expertise.

This version of FANTASIA builds upon previous work from:

• Metazoa Phylogenomics Lab's FANTASIA
  The original implementation of FANTASIA for functional annotation.

• bio_embeddings
  A state-of-the-art framework for generating protein sequence embeddings.

• GoPredSim
  A similarity-based approach for Gene Ontology annotation.

• protein-metamorphisms-is
  Serves as the reference biological information system, providing a robust data model and curated datasets for protein structural and functional analysis.

We also extend our gratitude to LifeHUB-CSIC for inspiring this initiative and fostering innovation in computational biology.

Citing FANTASIA

If you use FANTASIA in your research, please cite the following publications:

1. Martínez-Redondo, G. I., Barrios, I., Vázquez-Valls, M., Rojas, A. M., & Fernández, R. (2024).
   Illuminating the functional landscape of the dark proteome across the Animal Tree of Life.
   DOI: 10.1101/2024.02.28.582465

2. Barrios-Núñez, I., Martínez-Redondo, G. I., Medina-Burgos, P., Cases, I., Fernández, R., & Rojas, A. M. (2024).
   Decoding proteome functional information in model organisms using protein language models.
   DOI: 10.1101/2024.02.14.580341

---

👥 Project Team

🔧 Technical Team

• Francisco Miguel Pérez Canales: [e-mail]
  Author of the system’s engineering and technical implementation
• Francisco J. Ruiz Mota: [e-mail]
  Junior developer

🧬 Scientific Team & Original Authors of FANTASIA v1

• Ana M. Rojas: [e-mail]
• Gemma I. Martínez-Redondo: [e-mail]
• Rosa Fernández: [e-mail]

---