Albi Celaj
Research Subjects
Research Interests
Albi Celaj has a total of 125 co-authors. Here are the top co-authors by number of shared publications:
- Christopher Bernard Cole
Deep Genomics
5 shared publications Helen Zhu
Deep Genomics Inc., Toronto, Ontario, Canada
5 shared publications- Brendan J. Frey
Deep Genomics
4 shared publications - Bhargav Kanuparthi
Deep Genomics
3 shared publications Diane Ly
Deep Genomics
3 shared publicationsKen Kron
Deep Genomics
3 shared publications
Latest Publications
- Open Access
DeepRNAi: Predicting on- and off-target siRNA activity and establishing a benchmark
Small interfering RNAs (siRNAs) silence target genes through RISC-mediated hybridization and cleavage, but can silence unintended transcripts and cause toxicity or complicate regulatory approval. In drug development, empirical assessment by qPCR and RNA-seq is slow, forces early candidate prioritization, and is limited by available species- and genome-appropriate cell models. While extensive empirical screening remains standard practice, there is a significant opportunity for modern AI methods to change this paradigm. We introduce DeepRNAi, a deep learning model that accurately predicts on- and off-target knockdown from siRNA and mRNA sequence. Trained on 16 million siRNA-transcript pairs from 2,031 siRNAs designed to interrogate known siRNA failure modes, and incorporating transcript context from biological foundation models, DeepRNAi is the first in silico method that approaches experimental accuracy. It generalizes across cell models, transcripts, genomes and species, and in a small study distinguishes approved therapeutics from hepatotoxic candidates. The value of DeepRNAi for prioritizing better candidates up front improving likelihood of success later in development is evidenced by a genome-wide survey of 24.4 million siRNAs targeting 7,946 genes, which indicates that 90.1% of the 100 most potent siRNAs have at least some off-target concerns. To support the community, we are publishing a training and benchmark dataset containing 1.84 million siRNA-transcript pairs from 240 siRNAs.
Open AccessDeepRNAi: Predicting on- and off-target siRNA activity and establishing a benchmark
Small interfering RNAs (siRNAs) silence target genes through RISC-mediated hybridization and cleavage, but can silence unintended transcripts and cause toxicity or complicate regulatory approval. In drug development, empirical assessment by qPCR and RNA-seq is slow, forces early candidate prioritization, and is limited by available species- and genome-appropriate cell models. While extensive empirical screening remains standard practice, there is a significant opportunity for modern AI methods to change this paradigm. We introduce DeepRNAi, a deep learning model that accurately predicts on- and off-target knockdown from siRNA and mRNA sequence. Trained on 16 million siRNA-transcript pairs from 2,031 siRNAs designed to interrogate known siRNA failure modes, and incorporating transcript context from biological foundation models, DeepRNAi is the first in silico method that approaches experimental accuracy. It generalizes across cell models, transcripts, genomes and species, and in a small study distinguishes approved therapeutics from hepatotoxic candidates. The value of DeepRNAi for prioritizing better candidates up front improving likelihood of success later in development is evidenced by a genome-wide survey of 24.4 million siRNAs targeting 7,946 genes, which indicates that 90.1% of the 100 most potent siRNAs have at least some off-target concerns. To support the community, we are publishing a training and benchmark dataset containing 1.84 million siRNA-transcript pairs from 240 siRNAs.
- Open Access
Enigma: An Efficient Model for Deciphering Regulatory Genomics
Genomic sequence-to-function models have emerged as powerful tools for deciphering cis-regulatory grammar to advance our understanding of disease biology and guide therapeutic development. Recent advances have been driven by multi-task training of large transformer-based models on thousands of genome tracks. However, these performance gains have come at significant computational cost for both training and inference, hindering large-scale applications and slowing future model development. Here, rather than continuing to scale model size and add more training tracks, we focus on architectural efficiency and train on a substantially smaller, curated set of genome tracks. Our model Enigma achieves competitive performance with current state-of-the-art models at single-base resolution while substantially reducing computational cost. On zero-shot variant effect prediction benchmarks, Enigma outperforms the leading open-source model, the Borzoi ensemble, while using 10.9% of its compute and improving resolution from 32 bases to a single base. Compared to AlphaGenome, Enigma achieves 90.4 - 97.3% of its performance using 7.5% of its estimated compute. These improvements in efficiency can facilitate further development of models for regulatory genomics. We demonstrate this by fine-tuning Enigma on predicting three new molecular phenotypes - ChIP-seq, RNA half-life, and translation efficiency - achieving or exceeding the performance of state-of-the-art task-specific models. We are providing Enigma for non-commercial use to benefit the broader research field.
Open AccessEnigma: An Efficient Model for Deciphering Regulatory Genomics
Genomic sequence-to-function models have emerged as powerful tools for deciphering cis-regulatory grammar to advance our understanding of disease biology and guide therapeutic development. Recent advances have been driven by multi-task training of large transformer-based models on thousands of genome tracks. However, these performance gains have come at significant computational cost for both training and inference, hindering large-scale applications and slowing future model development. Here, rather than continuing to scale model size and add more training tracks, we focus on architectural efficiency and train on a substantially smaller, curated set of genome tracks. Our model Enigma achieves competitive performance with current state-of-the-art models at single-base resolution while substantially reducing computational cost. On zero-shot variant effect prediction benchmarks, Enigma outperforms the leading open-source model, the Borzoi ensemble, while using 10.9% of its compute and improving resolution from 32 bases to a single base. Compared to AlphaGenome, Enigma achieves 90.4 - 97.3% of its performance using 7.5% of its estimated compute. These improvements in efficiency can facilitate further development of models for regulatory genomics. We demonstrate this by fine-tuning Enigma on predicting three new molecular phenotypes - ChIP-seq, RNA half-life, and translation efficiency - achieving or exceeding the performance of state-of-the-art task-specific models. We are providing Enigma for non-commercial use to benefit the broader research field.
- Open Access
FlashRNA: An Efficient Model for Regulatory Genomics
Transformer-based genomic sequence-to-function models effectively capture long-range genomic interactions but incur high computational costs due to the quadratic complexity of their self-attention layers. In this work, we introduce FlashRNA , which significantly improves computational and memory efficiency through FlashAttention , advancements in model architecture, and optimized training setup. FlashRNA achieves comparable or slightly improved predictive performance compared to similar sized Borzoi or Flashzoi models, notably without depending on pre-trained weights – a major limitation of Flashzoi . Remarkably, we trained FlashRNA from scratch in one day on a single GPU, significantly accelerating training and inference speed. These improvements can facilitate further developments in models for regulatory genomics by reducing computational cost. We demonstrate this in two downstream applications: 1) we train a large ensemble of 16 FlashRNA models and distill them into a single model to improve performance while maintaining efficiency, and 2) we fine-tune FlashRNA on three prediction tasks – ChIP-seq, RNA half-life, and translation efficiency – achieving performance matching or exceeding state-of-the-art task-specific models. Code: https://github.com/deepgenomics/flashrna
Open AccessFlashRNA: An Efficient Model for Regulatory Genomics
Transformer-based genomic sequence-to-function models effectively capture long-range genomic interactions but incur high computational costs due to the quadratic complexity of their self-attention layers. In this work, we introduce FlashRNA , which significantly improves computational and memory efficiency through FlashAttention , advancements in model architecture, and optimized training setup. FlashRNA achieves comparable or slightly improved predictive performance compared to similar sized Borzoi or Flashzoi models, notably without depending on pre-trained weights – a major limitation of Flashzoi . Remarkably, we trained FlashRNA from scratch in one day on a single GPU, significantly accelerating training and inference speed. These improvements can facilitate further developments in models for regulatory genomics by reducing computational cost. We demonstrate this in two downstream applications: 1) we train a large ensemble of 16 FlashRNA models and distill them into a single model to improve performance while maintaining efficiency, and 2) we fine-tune FlashRNA on three prediction tasks – ChIP-seq, RNA half-life, and translation efficiency – achieving performance matching or exceeding state-of-the-art task-specific models. Code: https://github.com/deepgenomics/flashrna
- Open Access
Sequence based prediction of cell type specific microRNA binding and mRNA degradation for therapeutic discovery
MicroRNAs and RNA binding proteins are crucial elements of post-transcriptional gene regulation, which governs the fate of mRNA molecules in the cell. However, the landscape of these regulatory interactions, particularly across different mammalian cell types, remains underexplored. We describe REPRESS, a deep learning model that predicts cell-type-specific microRNA binding and mRNA degradation directly from RNA sequence. REPRESS was trained on AGO2-CLIP, miR-eCLIP and Degradome-Seq data profiling millions of microRNA binding and mRNA degradation sites across multiple cell types in human and mouse. It reveals biology that other state-of-the-art methods did not, such as identifying repressive non-canonical miRNA target sites and decoding the regulatory effects of sequence context and miRNA binding site multiplicity. REPRESS outperforms other advanced methods and neural architectures on a comprehensive suite of seven orthogonal tasks, including identifying genetic variants that affect microRNA binding, predicting out-of-distribution data from massively parallel reporter assays, and predicting canonical and non-canonical miRNA mediated repression. To demonstrate the general utility of REPRESS, we show that it provides insights into novel biology and the design of RNA therapeutics. Code is available at : https://github.com/deepgenomics/repress
Open AccessSequence based prediction of cell type specific microRNA binding and mRNA degradation for therapeutic discovery
MicroRNAs and RNA binding proteins are crucial elements of post-transcriptional gene regulation, which governs the fate of mRNA molecules in the cell. However, the landscape of these regulatory interactions, particularly across different mammalian cell types, remains underexplored. We describe REPRESS, a deep learning model that predicts cell-type-specific microRNA binding and mRNA degradation directly from RNA sequence. REPRESS was trained on AGO2-CLIP, miR-eCLIP and Degradome-Seq data profiling millions of microRNA binding and mRNA degradation sites across multiple cell types in human and mouse. It reveals biology that other state-of-the-art methods did not, such as identifying repressive non-canonical miRNA target sites and decoding the regulatory effects of sequence context and miRNA binding site multiplicity. REPRESS outperforms other advanced methods and neural architectures on a comprehensive suite of seven orthogonal tasks, including identifying genetic variants that affect microRNA binding, predicting out-of-distribution data from massively parallel reporter assays, and predicting canonical and non-canonical miRNA mediated repression. To demonstrate the general utility of REPRESS, we show that it provides insights into novel biology and the design of RNA therapeutics. Code is available at : https://github.com/deepgenomics/repress
- Open Access
An RNA foundation model enables discovery of disease mechanisms and candidate therapeutics
Accurately modeling and predicting RNA biology has been a long-standing challenge, bearing significant clinical ramifications for variant interpretation and the formulation of tailored therapeutics. We describe a foundation model for RNA biology, “BigRNA”, which was trained on thousands of genome-matched datasets to predict tissue-specific RNA expression, splicing, microRNA sites, and RNA binding protein specificity from DNA sequence. Unlike approaches that are restricted to missense variants, BigRNA can identify pathogenic non-coding variant effects across diverse mechanisms, including polyadenylation, exon skipping and intron retention. BigRNA accurately predicted the effects of steric blocking oligonucleotides (SBOs) on increasing the expression of 4 out of 4 genes, and on splicing for 18 out of 18 exons across 14 genes, including those involved in Wilson disease and spinal muscular atrophy. We anticipate that BigRNA and foundation models like it will have widespread applications in the field of personalized RNA therapeutics.
Open AccessAn RNA foundation model enables discovery of disease mechanisms and candidate therapeutics
Accurately modeling and predicting RNA biology has been a long-standing challenge, bearing significant clinical ramifications for variant interpretation and the formulation of tailored therapeutics. We describe a foundation model for RNA biology, “BigRNA”, which was trained on thousands of genome-matched datasets to predict tissue-specific RNA expression, splicing, microRNA sites, and RNA binding protein specificity from DNA sequence. Unlike approaches that are restricted to missense variants, BigRNA can identify pathogenic non-coding variant effects across diverse mechanisms, including polyadenylation, exon skipping and intron retention. BigRNA accurately predicted the effects of steric blocking oligonucleotides (SBOs) on increasing the expression of 4 out of 4 genes, and on splicing for 18 out of 18 exons across 14 genes, including those involved in Wilson disease and spinal muscular atrophy. We anticipate that BigRNA and foundation models like it will have widespread applications in the field of personalized RNA therapeutics.
- Open Access
- 25 September 2023
- Published by Springer Nature in Research Square
An RNA foundation model enables discovery of disease mechanisms and candidate therapeutics
Accurately modeling and predicting RNA biology has been a long-standing challenge, bearing significant clinical ramifications for variant interpretation and the formulation of tailored therapeutics. We describe a foundation model for RNA biology, “BigRNA”, which was trained on thousands of genome-matched datasets to predict tissue-specific RNA expression, splicing, microRNA sites, and RNA binding protein specificity from DNA sequence. Unlike approaches that are restricted to missense variants, BigRNA can identify pathogenic non-coding variant effects across diverse mechanisms, including polyadenylation, exon skipping and intron retention. BigRNA accurately predicted the effects of steric blocking oligonucleotides (SBOs) on increasing the expression of 4 out of 4 genes, and on splicing for 18 out of 18 exons across 14 genes, including those involved in Wilson disease and spinal muscular atrophy. We anticipate that BigRNA and foundation models like it will have widespread applications in the field of personalized RNA therapeutics.
- 25 September 2023
- Published by Springer Nature in Research Square
Open AccessAn RNA foundation model enables discovery of disease mechanisms and candidate therapeutics
Accurately modeling and predicting RNA biology has been a long-standing challenge, bearing significant clinical ramifications for variant interpretation and the formulation of tailored therapeutics. We describe a foundation model for RNA biology, “BigRNA”, which was trained on thousands of genome-matched datasets to predict tissue-specific RNA expression, splicing, microRNA sites, and RNA binding protein specificity from DNA sequence. Unlike approaches that are restricted to missense variants, BigRNA can identify pathogenic non-coding variant effects across diverse mechanisms, including polyadenylation, exon skipping and intron retention. BigRNA accurately predicted the effects of steric blocking oligonucleotides (SBOs) on increasing the expression of 4 out of 4 genes, and on splicing for 18 out of 18 exons across 14 genes, including those involved in Wilson disease and spinal muscular atrophy. We anticipate that BigRNA and foundation models like it will have widespread applications in the field of personalized RNA therapeutics.
- Open Access
- 1 May 2018
- Published by Springer Nature in Molecular Systems Biology
- Vol. 14 (5) , e7985
Mapping DNA damage‐dependent genetic interactions in yeast via party mating and barcode fusion genetics
Condition‐dependent genetic interactions can reveal functional relationships between genes that are not evident under standard culture conditions. State‐of‐the‐art yeast genetic interaction mapping, which relies on robotic manipulation of arrays of double‐mutant strains, does not scale readily to multi‐condition studies. Here, we describe barcode fusion genetics to map genetic interactions (BFG‐GI), by which double‐mutant strains generated via en masse “party” mating can also be monitored en masse for growth to detect genetic interactions. By using site‐specific recombination to fuse two DNA barcodes, each representing a specific gene deletion, BFG‐GI enables multiplexed quantitative tracking of double mutants via next‐generation sequencing. We applied BFG‐GI to a matrix of DNA repair genes under nine different conditions, including methyl methanesulfonate (MMS), 4‐nitroquinoline 1‐oxide (4NQO), bleomycin, zeocin, and three other DNA‐damaging environments. BFG‐GI recapitulated known genetic interactions and yielded new condition‐dependent genetic interactions. We validated and further explored a subnetwork of condition‐dependent genetic interactions involving MAG1 , SLX4, and genes encoding the Shu complex, and inferred that loss of the Shu complex leads to an increase in the activation of the checkpoint protein kinase Rad53.
- 1 May 2018
- Published by Springer Nature in Molecular Systems Biology
- Vol. 14 (5) , e7985
Open AccessMapping DNA damage‐dependent genetic interactions in yeast via party mating and barcode fusion genetics
Condition‐dependent genetic interactions can reveal functional relationships between genes that are not evident under standard culture conditions. State‐of‐the‐art yeast genetic interaction mapping, which relies on robotic manipulation of arrays of double‐mutant strains, does not scale readily to multi‐condition studies. Here, we describe barcode fusion genetics to map genetic interactions (BFG‐GI), by which double‐mutant strains generated via en masse “party” mating can also be monitored en masse for growth to detect genetic interactions. By using site‐specific recombination to fuse two DNA barcodes, each representing a specific gene deletion, BFG‐GI enables multiplexed quantitative tracking of double mutants via next‐generation sequencing. We applied BFG‐GI to a matrix of DNA repair genes under nine different conditions, including methyl methanesulfonate (MMS), 4‐nitroquinoline 1‐oxide (4NQO), bleomycin, zeocin, and three other DNA‐damaging environments. BFG‐GI recapitulated known genetic interactions and yielded new condition‐dependent genetic interactions. We validated and further explored a subnetwork of condition‐dependent genetic interactions involving MAG1 , SLX4, and genes encoding the Shu complex, and inferred that loss of the Shu complex leads to an increase in the activation of the checkpoint protein kinase Rad53.
- Open Access
- 1 July 2017
- Published by Springer Nature in Molecular Systems Biology
- Vol. 13 (7) , 934
Quantitative analysis of protein interaction network dynamics in yeast
Many cellular functions are mediated by protein–protein interaction networks, which are environment dependent. However, systematic measurement of interactions in diverse environments is required to better understand the relative importance of different mechanisms underlying network dynamics. To investigate environment‐dependent protein complex dynamics, we used a DNA‐barcode‐based multiplexed protein interaction assay in Saccharomyces cerevisiae to measure in vivo abundance of 1,379 binary protein complexes under 14 environments. Many binary complexes (55%) were environment dependent, especially those involving transmembrane transporters. We observed many concerted changes around highly connected proteins, and overall network dynamics suggested that “concerted” protein‐centered changes are prevalent. Under a diauxic shift in carbon source from glucose to ethanol, a mass‐action‐based model using relative mRNA levels explained an estimated 47% of the observed variance in binary complex abundance and predicted the direction of concerted binary complex changes with 88% accuracy. Thus, we provide a resource of yeast protein interaction measurements across diverse environments and illustrate the value of this resource in revealing mechanisms of network dynamics.
- 1 July 2017
- Published by Springer Nature in Molecular Systems Biology
- Vol. 13 (7) , 934
Open AccessQuantitative analysis of protein interaction network dynamics in yeast
Many cellular functions are mediated by protein–protein interaction networks, which are environment dependent. However, systematic measurement of interactions in diverse environments is required to better understand the relative importance of different mechanisms underlying network dynamics. To investigate environment‐dependent protein complex dynamics, we used a DNA‐barcode‐based multiplexed protein interaction assay in Saccharomyces cerevisiae to measure in vivo abundance of 1,379 binary protein complexes under 14 environments. Many binary complexes (55%) were environment dependent, especially those involving transmembrane transporters. We observed many concerted changes around highly connected proteins, and overall network dynamics suggested that “concerted” protein‐centered changes are prevalent. Under a diauxic shift in carbon source from glucose to ethanol, a mass‐action‐based model using relative mRNA levels explained an estimated 47% of the observed variance in binary complex abundance and predicted the direction of concerted binary complex changes with 88% accuracy. Thus, we provide a resource of yeast protein interaction measurements across diverse environments and illustrate the value of this resource in revealing mechanisms of network dynamics.