Accurate prediction of virus-host protein-protein interactions via a Siamese neural network using deep protein sequence embeddings

Madan, Sumit; Demina, Victoria; Stapf, Marcus; Ernst, Oliver; Fröhlich, Holger

doi:10.1016/j.patter.2022.100551

Cited by 10 publications

(14 citation statements)

References 54 publications

(129 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We conducted a comparative analysis between SENSE-PPI and several existing sequence-based deep learning models, namely PIPR [4], D-SCRIPT [53], Topsy-Turvy [52] and STEP [28].…”

Section: Other Deep Learning Architectures Used For Comparisonmentioning

confidence: 99%

“…Comparative analysis of the performance of SENSE-PPI and several others DL architectures on Guo's yeast dataset. Comparison is made based on a 5-fold cross-validation test between SENSE-PPI, PIPR, STEP, and the original model from Guo et al Values reported for all DL architectures other than SENSE-PPI were taken from[28].The best values across architectures are shown in bold.of sequence pairs at each cross-validating fold, in the validation step, each model mostly evaluates proteins already "seen" during training. Such pairs are known to show better performance in comparison to those excluded in training[35,13].3.2 Training on the human proteome and testing on model and non-model organismsThe recent global e↵orts to sequence the biodiversity of species[60,3,23,24,41], make PPI predictions in yet unexplored organisms a major challenge.…”

mentioning

confidence: 99%

See 1 more Smart Citation

SENSE-PPI reconstructs protein-protein interactions of various complexities, within, across, and between species, with sequence-based evolutionary scale modeling and deep learning

Volzhenin,

Bittner,

Carbone

2023

Preprint

View full text Add to dashboard Cite

Ab initio computational reconstructions of protein-protein interaction networks will bring invaluable information on cellular systems leading to the discovery of potential molecular interactions and to the unraveling of new biological mechanisms. Crossed with large-scale experimental data, they will reduce the complexity of the set of proteins involved in a biological mechanism that should be experimentally tested. The high confidence of the network is a prerequisite of these advancements. We combine advances in protein language models and recurrent neural networks to recover interaction signals in datasets of biological sequences with increasing complexity. SENSE-PPI (Sequence-based EvolutioNary ScalE Protein-Protein Interaction) is a deep learning model that predicts physical interactions between pairs of proteins using only their amino acid sequences. SENSE-PPI is highly accurate, demands limited training, keeps high accuracy levels when trained on one species and validated on another, possibly a non-model organism, and it can predict protein interactions between human and viruses. It determines protein partners that might have known structures or be intrinsically disordered, it reconstructs ab initio protein interaction networks by discriminating partners among tens of thousands of proteins and identifies partners between sets of proteins sharing similar functions, like enzymes vs inhibitors. SENSE-PPI is very fast and can test a few hundred proteins against themselves in less than a few minutes. SENSE-PPI learns better than existing deep learning methods addressing the problem of PPI reconstruction and it generalizes to different related problems, such as human-viral and intrinsically disordered protein interaction. A comparative analysis of the sequence embeddings generated by different versions of ESM2 in SENSE-PPI leads to recognizing the importance of a large number of parameters in protein language models used for PPI network reconstruction. The user will have access to two versions of SENSE-PPI one favoring precision and the other recall, as well as to a newly designed dataset of almost 1 million positive and negative human interactions to use for testing and validation.

show abstract

“…We conducted a comparative analysis between SENSE-PPI and several existing sequence-based deep learning models, namely PIPR [4], D-SCRIPT [53], Topsy-Turvy [52] and STEP [28].…”

Section: Other Deep Learning Architectures Used For Comparisonmentioning

confidence: 99%

mentioning

confidence: 99%

SENSE-PPI reconstructs protein-protein interactions of various complexities, within, across, and between species, with sequence-based evolutionary scale modeling and deep learning

Volzhenin,

Bittner,

Carbone

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Methods for such an approach include autoencoders, convolutional neural networks [3], or sequential models like recurrent neural networks (RNN) [4] or transformer-based models [5][6][7][8][9]. Transformer-based models originate from natural language processing (NLP) and have recently gained much attention since they have achieved excellent results in many areas [10][11][12][13][14]. A principal advantage of transformer models is the ability to train them in a parallel fashion and the ability to weigh different parts of a time series differently due to their inbuilt attention mechanism.…”

Section: Introductionmentioning

confidence: 99%

A Transformer-Based Model Trained on Large Scale Claims Data for Prediction of Severe COVID-19 Disease Progression

Lentzen

Lindén

Veeranki

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

In situations like the COVID-19 pandemic, healthcare systems are under enormous pressure as they can rapidly collapse under the burden of the crisis. Machine learning (ML) based risk models could lift the burden by identifying patients with high risk of severe disease progression. Electronic Health Records (EHRs) provide crucial sources of information to develop these models because they rely on routinely collected healthcare data. However, EHR data is challenging for training ML models because it contains irregularly timestamped diagnosis, prescription, and procedure codes. For such data, transformer-based models are promising. We extended the previously published Med-BERT model by including age, sex, medications, quantitative clinical measures, and state information. After pre-training on approximately 988 million EHRs from 3.5 million patients, we developed models to predict Acute Respiratory Manifestations (ARM) risk using the medical history of 80,211 COVID-19 patients. Compared to XGBoost and Random Forests, our transformer-based models more accurately forecast the risk of developing ARM after COVID-19 infection. We used Integrated Gradients and Bayesian networks to understand the link between the essential features of our model. Finally, we evaluated adapting our model to Austrian in-patient data. Our study highlights the promise of predictive transformer-based models for precision medicine.

show abstract

“…Most of the work in this area focuses on prediction of interactions from sequence, especially using deep learning techniques. Some recent publications reported highly accurate prediction results from sequence alone that caught our attention ( Tsukiyama et al, 2021 ; Asim et al, 2022 ; Madan et al, 2022 ). As long-time practiotioners of machine learning in this area, we approach such results with a healthy dose of skepticism.…”

Section: Introductionmentioning

confidence: 99%

“…Although some researchers have rightfully shunned the technique of similarity-constrained negative example selection ( Liu-Wei et al, 2021 ; Madan et al, 2022 ), this practice remains present in the field of HPI prediction ( Basit et al, 2018 ; Zhou et al, 2018 ; Yang et al, 2020 ; Pitta et al, 2021 ; Tsukiyama et al, 2021 ; Yang et al, 2021 ; Asim et al, 2022 ) and also in PPI prediction Chen et al (2022) , necessitating this paper to alert researchers to this issue. We have also observed the use of similarity based choice of negative examples in other sequence-based prediction problems such as anti-microbial peptide prediction ( Veltri et al, 2018 ).…”

Section: Introductionmentioning

confidence: 99%

On the choice of negative examples for prediction of host-pathogen protein interactions

et al. 2022

View full text Add to dashboard Cite

As practitioners of machine learning in the area of bioinformatics we know that the quality of the results crucially depends on the quality of our labeled data. While there is a tendency to focus on the quality of positive examples, the negative examples are equally as important. In this opinion paper we revisit the problem of choosing negative examples for the task of predicting protein-protein interactions, either among proteins of a given species or for host-pathogen interactions and describe important issues that are prevalent in the current literature. The challenge in creating datasets for this task is the noisy nature of the experimentally derived interactions and the lack of information on non-interacting proteins. A standard approach is to choose random pairs of non-interacting proteins as negative examples. Since the interactomes of all species are only partially known, this leads to a very small percentage of false negatives. This is especially true for host-pathogen interactions. To address this perceived issue, some researchers have chosen to select negative examples as pairs of proteins whose sequence similarity to the positive examples is sufficiently low. This clearly reduces the chance for false negatives, but also makes the problem much easier than it really is, leading to over-optimistic accuracy estimates. We demonstrate the effect of this form of bias using a selection of recent protein interaction prediction methods of varying complexity, and urge researchers to pay attention to the details of generating their datasets for potential biases like this.

show abstract

Accurate prediction of virus-host protein-protein interactions via a Siamese neural network using deep protein sequence embeddings

Cited by 10 publications

References 54 publications

SENSE-PPI reconstructs protein-protein interactions of various complexities, within, across, and between species, with sequence-based evolutionary scale modeling and deep learning

SENSE-PPI reconstructs protein-protein interactions of various complexities, within, across, and between species, with sequence-based evolutionary scale modeling and deep learning

A Transformer-Based Model Trained on Large Scale Claims Data for Prediction of Severe COVID-19 Disease Progression

On the choice of negative examples for prediction of host-pathogen protein interactions

Contact Info

Product

Resources

About