Pith. sign in

REVIEW 4 major objections 5 minor 63 references

CNN-LSTM Hybrid Model for AI-Driven Prediction of COVID-19 Severity from Spike Sequences and Clinical Data

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A hybrid CNN-LSTM model predicts COVID-19 severity from spike protein sequences and clinical metadata, reporting an F1 of 82.92% and ROC-AUC of 0.9084 on South American cases.

desk verdict The headline metrics contradict the paper's own confusion matrix, and the feature descriptions conflict; this paper is not referee-ready. read the letter →

arxiv 2505.23879 v1 pith:2HMMLTSC submitted 2025-05-29 cs.LG cs.AI

Caio Cheohen , Vinnícius M. S. Gomes , Manuela L. da Silva This is my paper
classification cs.LGcs.AI
keywords CNN-LSTMhybridmodelCOVID-19severitypredictionspikeproteinsequenceclinicaloutcomeclassificationreceptor-bindingdomaingenomicsurveillancedeeplearningSouthAmerica
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that a hybrid CNN-LSTM network can predict whether a COVID-19 patient's illness will be mild or severe from the SARS-CoV-2 spike protein sequence combined with basic clinical metadata. On 3,467 standardized South American records (2,313 severe, 1,154 mild) drawn from a public genomic repository, the model reaches an F1 of 82.92%, precision of 83.56%, recall of 82.85%, and ROC-AUC of 0.9084. The authors see this as evidence that amino-acid-level sequence information, especially in the receptor-binding domain, carries a usable severity signal even without explicit structural or immunological features. If the result holds, it gives genomic surveillance a practical route to early severity stratification in future outbreaks.

What carries the argument

The load-bearing machinery is the hybrid CNN-LSTM stack plus the feature encoding. Four 1D convolutional layers with kernel size 4 scan the 3,013-element padded sequence vectors to pick out local motifs; max pooling and dropout reduce overfitting; a 64-unit LSTM layer then models dependencies across the whole sequence; dense layers with a sigmoid output produce the mild/severe probability. The input vector packs amino acid composition, sequence length, diversity, mean hydrophobicity, net charge, predicted secondary-structure fractions, polarity, hydrogen-bonding propensity, and ten-dimensional per-residue encodings in which residues in the receptor-binding domain (positions 319–541) receive a weight of 5, plus one-hot clinical and demographic variables. Synthetic oversampling balances the training classes.

What would settle it

An independent validation on a cohort with prospectively recorded, standardized clinical outcomes would settle it: if F1 and ROC-AUC do not stay near 82.9% and 0.9084, the reported signal does not generalize; a quicker check is retraining with Mild/Severe labels randomly permuted, which should drop performance to chance.

Watch

Extended reading notes

Core claim

The central claim is that a CNN-LSTM architecture trained on a padded numeric encoding of spike protein sequences—physicochemical descriptors, receptor-binding-domain-focused residue weights, and one-hot demographic and lineage/clade variables—can separate mild from severe COVID-19 outcomes. The paper reports 83.56% precision, 82.85% recall, 82.92% F1, and 0.9084 ROC-AUC on a held-out test set, with training accuracy stabilizing around 85% and little evidence of overfitting. It also reports that the dominant lineages (P.1, AY.99.2) and clades (GR, GK) match regional epidemiological trends, which the authors take as consistent with viral genetics influencing clinical outcome. The paper does not assert a mechanism for severity; the claim is empirical and tied to this dataset and feature encoding.

Load-bearing premise

The model's numbers depend on the free-text clinical outcomes in the source repository being mapped accurately into Mild and Severe labels, since the mapping drops about two thirds of the samples and groups death and ICU care under 'severe' and home care or asymptomatic infection under 'mild'.

Editorial extensions

If this is right

  • Spike sequences collected early in an infection could be screened to flag patients at higher risk of severe COVID-19, supporting earlier intervention and resource allocation.
  • The reported performance suggests that sequence-derived physicochemical descriptors plus a CNN-LSTM combination can capture clinically relevant signal without needing structural or immunological measurements.
  • The paper's free-text-to-status standardization provides a template for cleaning clinical outcome metadata in genomic repositories for future outbreak analytics.
  • The model's lineage and clade distribution matching regional epidemiological trends means the severity signal should be re-evaluated with lineage and collection date controlled before attributing it to specific mutations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reported AUC may partly reflect lineage and calendar-time structure in the data, since P.1 and AY.99.2 dominated different epidemic waves and hospital admission policies changed over time; a within-lineage, within-period analysis would separate viral-pathogenicity signal from era effects.
  • Because the labels come from hospitalization and death records, the model may be learning health-system contact and reporting practices as much as biological severity; validation on an independent clinical severity scale would distinguish these.
  • The 64% sample reduction from metadata standardization could bias the cohort toward submitters with complete records; reproducing the pipeline on a prospectively collected, standardized-outcome cohort would clarify generalizability.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper reports a CNN-LSTM hybrid model for predicting COVID-19 severity from SARS-CoV-2 spike protein sequences and clinical metadata of South American patients. A GISAID-derived dataset of 3,467 samples (2,313 severe, 1,154 mild) is used, with features from sequence encoding, physicochemical descriptors, RBD-focused position weighting, and one-hot encoded demographic and viral lineage/clade variables. The authors report strong test performance (F1 82.92%, ROC-AUC 0.9084, precision 83.56%, recall 82.85%) and conclude that sequence-based deep learning can stratify mild from severe cases. The contribution is empirical and data-dependent; no mechanistic or theoretical derivation is attempted.

Significance. If the reported performance were reproducible and the evaluation sound, the paper would be a useful proof-of-concept for genomic surveillance and severity prediction. The authors make their code publicly available and use a relevant regional dataset, which are positive features. However, the central performance claim is currently unsupported because the published confusion matrix and the headline metrics are arithmetically inconsistent. Additional ambiguities about SMOTE placement, label construction, and contradictory descriptions of the model inputs further weaken the empirical evidence. The paper's main value would therefore depend on a careful correction and full re-evaluation of the reported results.

major comments (4)
  1. [Section 3.2, Tables 3 and 4] The precision, recall, and F1 scores in Table 4 cannot be derived from the confusion matrix in Table 3 under either class convention. If the positive class is Mild, as Section 2.6 states (mild=1), the matrix gives TP=190, FP=84, FN=37, TN=383, yielding precision=190/(190+84)=0.693, recall=190/(190+37)=0.837, and F1=0.759. If the positive class is Severe, the corresponding values are precision=383/(383+37)=0.912, recall=383/(383+84)=0.820, and F1=0.863. Neither matches the reported precision 0.8356, recall 0.8285, or F1 0.8292. Notably, the reported sensitivity 0.8370 and specificity 0.8201 do match the matrix under the Mild-positive convention, which means at least one of the tables is wrong. Because the headline claim of the paper is the F1/AUC performance, this internal inconsistency is a load-bearing error that must be resolved, with the correct numbers verified against the released code.
  2. [Section 2.8] The text states that the dataset was split into 80% training and 20% testing subsets and then balanced using SMOTE, but it does not specify whether SMOTE was applied before or after the split. If SMOTE is applied to the full dataset before splitting, synthetic examples would leak into the test set, making the reported metrics invalid. The authors must clarify the exact sequence of operations and confirm that SMOTE was applied only to the training partition, with the test set left in its original distribution.
  3. [Section 3.1, Table 1] The severity labels are derived from a subjective mapping of heterogeneous free-text clinical statuses, and all hospitalized-alive and moderate cases are assigned to an excluded 'Inconclusive' category. This mapping reduces the dataset by approximately 64% and restricts the prediction task to extreme outcomes (death/ICU versus home care/asymptomatic), which may inflate discriminability relative to a clinically realistic mild-versus-severe contrast. Given that the central claim rests entirely on this binary label, the authors should provide a sensitivity analysis using alternative label mappings (for example, treating any hospitalization as severe) and report the distribution and characteristics of the excluded samples.
  4. [Sections 2.2-2.7, 4, 5] There is a direct contradiction about what the model actually uses as input. The Methods describe the integration of global physicochemical descriptors, RBD-focused position-specific weighting, and one-hot encoded demographic and clinical variables, including clade and lineage (Sections 2.2-2.7). The Discussion and Conclusion, however, claim that the model was 'trained solely on spike protein sequences without incorporating structural or immunological covariates.' These statements cannot both be true, and the discrepancy changes the interpretation of the result: the reported AUC may reflect clinical metadata or lineage information rather than spike-sequence information. The authors must correct the description and, ideally, provide an ablation study separating the contribution of sequence-derived features from that of clinical and lineage covariates.
minor comments (5)
  1. [Section 3.2] The confusion matrix in Table 3 sums to 694, but 20% of 3,467 is 693.4; please reconcile the test-set size and explain the rounding.
  2. [Figure 4] The training curves show only accuracy and loss on the training set, so the claim of 'minimal overfitting' is not directly supported; a validation curve or a comparison of train and test metrics at the selected epoch would be more informative.
  3. [Section 2.4] Please clarify what the position-specific weight of 5 for RBD residues means in practice: whether it is a feature-scale multiplier, a sampling weight, or part of the loss function, and how it interacts with the one-hot secondary structure encoding.
  4. [Section 2.1, Table 1] The entries 'deceased 14/8' and 'deceased 20/8' are unclear; please spell out the intended dates and confirm whether they refer to the date of death.
  5. [Section 4] The cross-validation F1 of 84.87% and final test F1 of 82.92% are reported without variance or repeated-run information; a single split does not support strong claims about generalization, so reporting mean and standard deviation over multiple random seeds would be preferable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper reports a standard empirical train/test evaluation, and its supporting claims do not reduce to their inputs.

full rationale

The paper's central claim is an empirical performance result (F1=82.92%, ROC-AUC=0.9084) obtained by training a CNN-LSTM classifier on spike-sequence features plus demographic and viral-category covariates and evaluating on a held-out 20% test split (Sections 2.8 and 3.2). No derived quantity is defined in terms of the predicted outcome, no fitted parameter is renamed as a prediction, and no uniqueness or ansatz result is imported from prior work by the same authors. The single self-citation [54] (Dias et al., which includes author M. L. da Silva) supports only the background statement that P.1 and GR/GK lineages match regional epidemiological trends; it is not load-bearing for the model architecture, feature construction, or test metrics. The apparent mismatch between the Table 3 confusion matrix and Table 4 precision/recall/F1 values, and the tension between the Methods' use of clinical metadata and the Discussion's claim of training 'solely on spike protein sequences,' are internal consistency and correctness concerns, not circular reasoning. The evaluation is self-contained against an external data source (GISAID) and reports held-out metrics, so the appropriate circularity finding is none.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central contribution is an empirical classifier, so no physical or mathematical derivation is claimed. The ledger lists the main fitted weights, tuned hyperparameters, hand-set feature weights, and domain assumptions that the reported performance depends on. The paper postulates no new particles, mediators, dimensions, or other entities.

free parameters (5)
  • Neural network weights (85,657 trainable parameters) = Not reported
    All trainable parameters of the CNN-LSTM are optimized against the training set during fitting; final weight values are not provided.
  • Optuna-selected architecture hyperparameters = Conv filters 128/64/64/24, kernel size 4, LSTM units 64, dropout 0.166
    Selected by Bayesian search on validation data; these choices directly shape the reported metrics.
  • RBD position weighting factor = 5 for residues 319 to 541, 1 elsewhere
    Hand-chosen weighting factor emphasizing the receptor-binding domain; not derived from data or a prior benchmark.
  • L2 regularization coefficient = 0.001
    Hand-set to prevent overfitting; affects training dynamics and reported generalization.
  • Sequence padding length = 3013
    All sequence representations are padded to the maximum observed length in this dataset; the input dimension is dataset-specific.
assumptions (4)
  • domain assumption GISAID clinical free-text entries can be reliably mapped to Mild/Severe with the excluded Inconclusive class.
    Used in Section 2.1 and Table 1; 64% of downloaded sequences are dropped during this mapping, so label quality and selection are load-bearing.
  • domain assumption Physicochemical descriptors, amino acid composition, and RBD-focused weighting capture outcome-relevant signal in spike sequences.
    Used in Sections 2.3 and 2.4; these features are proxies for protein behavior, not validated causal markers of severity.
  • domain assumption CNN local filters and LSTM long-range dependencies are an appropriate inductive bias for spike protein sequences.
    Used in Section 2.8; no comparison with other architectures is provided to support this assumption.
  • domain assumption The 80/20 split and SMOTE balancing produce a test set representative of future cases.
    Used in Section 2.8; time, geography, lineage, and ascertainment confounds are not adjusted, and SMOTE placement is ambiguous.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CNN-LSTM Hybrid Model for AI-Driven Prediction of COVID-19 Severity from Spike Sequences and Clinical Data." pith.science (2026). https://pith.science/paper/2HMMLTSC

@misc{pith2026250523879,
  author       = {Pith},
  title        = {Pith review of: CNN-LSTM Hybrid Model for AI-Driven Prediction of COVID-19 Severity from Spike Sequences and Clinical Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HMMLTSC}},
  note         = {Machine review of arXiv:2505.23879}
}
read the original abstract

The COVID-19 pandemic, caused by SARS-CoV-2, highlighted the critical need for accurate prediction of disease severity to optimize healthcare resource allocation and patient management. The spike protein, which facilitates viral entry into host cells, exhibits high mutation rates, particularly in the receptor-binding domain, influencing viral pathogenicity. Artificial intelligence approaches, such as deep learning, offer promising solutions for leveraging genomic and clinical data to predict disease outcomes. Objective: This study aimed to develop a hybrid CNN-LSTM deep learning model to predict COVID-19 severity using spike protein sequences and associated clinical metadata from South American patients. Methods: We retrieved 9,570 spike protein sequences from the GISAID database, of which 3,467 met inclusion criteria after standardization. The dataset included 2,313 severe and 1,154 mild cases. A feature engineering pipeline extracted features from sequences, while demographic and clinical variables were one-hot encoded. A hybrid CNN-LSTM architecture was trained, combining CNN layers for local pattern extraction and an LSTM layer for long-term dependency modeling. Results: The model achieved an F1 score of 82.92%, ROC-AUC of 0.9084, precision of 83.56%, and recall of 82.85%, demonstrating robust classification performance. Training stabilized at 85% accuracy with minimal overfitting. The most prevalent lineages (P.1, AY.99.2) and clades (GR, GK) aligned with regional epidemiological trends, suggesting potential associations between viral genetics and clinical outcomes. Conclusion: The CNN-LSTM hybrid model effectively predicted COVID-19 severity using spike protein sequences and clinical data, highlighting the utility of AI in genomic surveillance and precision public health. Despite limitations, this approach provides a framework for early severity prediction in future outbreaks.

Figures

Figures reproduced from arXiv: 2505.23879 by the authors.

Figure 1
Figure 1. CNN-LSTM Model Architecture – Overview of the CNN-LSTM model used for prediction. The architecture includes three 1D convolutional layers with max pooling and dropout, followed by an LSTM layer and a dense layer to produce the final predicted value. 4 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. SARS-CoV-2 Spike Protein Samples from South America – Geographic distribution of samples used in the training set across South American countries. The color scale from green to yellow represents the number of samples on a logarithmic scale, highlighting variation between countries. Areas shaded with gray hatching indicate countries with no available data. The absolute number of samples per country is as follows: Bra… view at source ↗
Figure 3
Figure 3. Top 20 Lineages found in South America samples: Retrieved lineages from January to March 2023 at the GISAID database after standardization process. 3.2 Model Training & Performance Metrics A confusion matrix is an essential tool used to evaluate the performance of binary classifiers by summarizing how often predicted labels correspond to actual labels. For a binary classification problem, where the true labels can b… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Training Performance of the Deep Learning Model Across Epochs – A: The increase of accuracy over epochs. B: The decrease of loss over epochs. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 60 canonical work pages

  1. [1]

    Outbreak of pneumonia of unknown etiology in wuhan, china: The mystery and the miracle

    Hongzhou Lu, Charles W Stratton, and Yi-Wei Tang. Outbreak of pneumonia of unknown etiology in wuhan, china: The mystery and the miracle. Journal of Medical Virology, 92(4):401–402, 2020

    work page 2020

  2. [2]

    A predictive model to explore risk factors for severe covid-19

    Feng-Hua Qian, Yang Cao, Yi-Xin Liu, Jing Huang, and Rong-Hua Zhu. A predictive model to explore risk factors for severe covid-19. Scientific Reports, 14:18197, 2024

    work page 2024

  3. [3]

    Predicting the disease outcome in covid-19 positive patients through machine learning: A retrospective cohort study with brazilian data

    Fábio SH De Souza, Natália S Hojo-Souza, Erick B Dos Santos, Christiane M Da Silva, and Daniel L Guidoni. Predicting the disease outcome in covid-19 positive patients through machine learning: A retrospective cohort study with brazilian data. Frontiers in Artificial Intelligence, 4:579931, 2021

    work page 2021

  4. [4]

    Human sars cov-2 spike protein mutations

    L Guruprasad. Human sars cov-2 spike protein mutations. Proteins: Structure, Function, and Bioinformatics, 89(5):569–576, 2021

    work page 2021

  5. [5]

    Sensitivity of two sars-cov-2 variants with spike protein mutations to neutralising antibodies

    Katharina Müller, Philipp Girl, Heiner von Buttlar, Gerhard Dobler, Roman Wölfel, Reinhold Förster, Katharina Esser-Nobis, V olker Lohmann, Thomas Günther, Georg Acker, et al. Sensitivity of two sars-cov-2 variants with spike protein mutations to neutralising antibodies. Virus Genes, 57(6):502–509, 2021

    work page 2021

  6. [6]

    Predicting natural evolution in the rbd region of the spike glycoprotein of sars-cov-2 by machine learning

    Yang Liu et al. Predicting natural evolution in the rbd region of the spike glycoprotein of sars-cov-2 by machine learning. Viruses, 16(1), 2024

    work page 2024

  7. [7]

    Empowering open data sharing for social good: a privacy-aware approach

    Tiago Carvalho, Luis Antunes, Carlos Costa Santos, and Nuno Moniz. Empowering open data sharing for social good: a privacy-aware approach. Scientific Data, 12:248, 2025

    work page 2025

  8. [8]

    Gisaid’s role in pandemic response

    Sangeeta Khare et al. Gisaid’s role in pandemic response. China CDC Weekly, 3(49):1049–1051, 2021

    work page 2021

Show all 63 references
  1. [9]

    Data, disease and diplomacy: Gisaid’s innovative contribution to global health

    Stefan Elbe and Gemma Buckland-Merrett. Data, disease and diplomacy: Gisaid’s innovative contribution to global health. Global Challenges, 1(1):33–46, 2017

    2017

  2. [10]

    An innovative ai-based primer design tool for precise and accurate detection of sars-cov-2 variants of concern

    Cesar A Perez-Romero et al. An innovative ai-based primer design tool for precise and accurate detection of sars-cov-2 variants of concern. Scientific Reports, 13:15782, 2023

    2023

  3. [11]

    Explainable artificial intelligence approaches for covid-19 prognosis prediction using clinical markers

    Krishnaraj Chadaga et al. Explainable artificial intelligence approaches for covid-19 prognosis prediction using clinical markers. Scientific Reports, 14:1783, 2024

    2024

  4. [12]

    Severity prediction for covid-19 patients via recurrent neural networks

    Joon Lee, Casey Ta, Judy H Kim, Chang Liu, and Chunhua Weng. Severity prediction for covid-19 patients via recurrent neural networks. AMIA Annual Symposium Proceedings, 2021:374–383, 2021

    2021

  5. [13]

    Covid-19 health data prediction: a critical evaluation of cnn-based approaches

    Tae Hyun Kim, Ramesh Chinthaginjala, Aravind Srinivasulu, Sai Prasanth Tera, and Saeed Osman Rab. Covid-19 health data prediction: a critical evaluation of cnn-based approaches. Scientific Reports, 15:9121, 2025

    2025

  6. [14]

    Cnn-lstm deep learning based forecasting model for covid-19 infection cases in nigeria, south africa and botswana

    Lawan Jibril Muhammad, Adamu Abubakar Haruna, Usman Sa’id Sharif, and Musa Baba Mohammed. Cnn-lstm deep learning based forecasting model for covid-19 infection cases in nigeria, south africa and botswana. Health Technology, 12(6):1259–1276, 2022

    2022

  7. [15]

    A machine learning model for the prediction of covid-19 severity using rna-seq, clinical, and co-morbidity data

    Sumit Sethi, Sounak Shakyawar, Anilkumar S Reddy, Jay C Patel, and Chittibabu Guda. A machine learning model for the prediction of covid-19 severity using rna-seq, clinical, and co-morbidity data. Diagnostics, 14(3), 2024

    2024

  8. [16]

    Predicting effects of noncoding variants with deep learning-based sequence model

    Jian Zhou and Olga G Troyanskaya. Predicting effects of noncoding variants with deep learning-based sequence model. Nature Methods, 12(10):931–934, 2015

    2015

  9. [17]

    Deep learning in bioinformatics: Introduction, application, and perspective in the big data era.Methods, 166:4–21, 2019

    Yu Li et al. Deep learning in bioinformatics: Introduction, application, and perspective in the big data era.Methods, 166:4–21, 2019

    2019

  10. [18]

    Deep learning in bioinformatics

    Seonwoo Min, Byunghan Lee, and Sungroh Yoon. Deep learning in bioinformatics. Briefings in Bioinformatics, 18(5):851–869, 2017

    2017

  11. [19]

    The relation between the divergence of sequence and structure in proteins

    Cyrus Chothia and Arthur M Lesk. The relation between the divergence of sequence and structure in proteins. The EMBO Journal, 5(4):823–826, 1986

    1986

  12. [20]

    Introduction to protein structure

    Anders Jacobsen et al. Introduction to protein structure. arXiv preprint arXiv:2307.02169, 2023

    2023 arXiv

  13. [21]

    Fundamentals of recurrent neural network (rnn) and long short-term memory (lstm) network

    Alex Sherstinsky. Fundamentals of recurrent neural network (rnn) and long short-term memory (lstm) network. Physica D: Nonlinear Phenomena, 404:132306, 2020

    2020

  14. [22]

    Recurrent neural networks and long short-term memory networks: Tutorial and survey

    Benyamin Ghojogh and Ali Ghodsi. Recurrent neural networks and long short-term memory networks: Tutorial and survey. arXiv preprint arXiv:2304.11461, 2023

    2023 arXiv

  15. [23]

    A high performance hybrid lstm cnn secure architecture for iot environments using deep learning

    Priyanka Sinha et al. A high performance hybrid lstm cnn secure architecture for iot environments using deep learning. Scientific Reports, 15:9684, 2025. 10 CNN-lSTM Hybrid Model For AI-Driven Prediction Of COVID-19 Severity From Spike Sequences And Clinical Data

    2025

  16. [24]

    High rate of mutational events in SARS-CoV-2 genomes across brazilian geographical regions, february 2020 to june 2021

    Ueric José Borges de Souza, Raíssa Nunes Dos Santos, Fabrício Souza Campos, Karine Lima Lourenço, Flavio Guimarães da Fonseca, Fernando Rosado Spilki, and Corona Ômica Br/McTi Network. High rate of mutational events in SARS-CoV-2 genomes across brazilian geographical regions, ...

    2020

  17. [25]

    A simple method for displaying the hydropathic character of a protein

    Jack Kyte and Russell F Doolittle. A simple method for displaying the hydropathic character of a protein. Journal of Molecular Biology, 157(1):105–132, 1982

    1982

  18. [26]

    Physiology, acid base balance

    Emily Hopkins, Terrence Sanvictores, and Sandeep Sharma. Physiology, acid base balance. StatPearls, 2025

    2025

  19. [27]

    Biopython: freely available python tools for computational molecular biology and bioinfor- matics

    Peter JA Cock et al. Biopython: freely available python tools for computational molecular biology and bioinfor- matics. Bioinformatics, 25(11):1422–1423, 2009

    2009

  20. [28]

    Prediction of protein antigenic determinants from amino acid sequences

    Thomas P Hopp and Kenneth R Woods. Prediction of protein antigenic determinants from amino acid sequences. Proceedings of the National Academy of Sciences, 78(6):3824–3828, 1981

    1981

  21. [29]

    Unraveling the structural and chemical features of biological short hydrogen bonds

    Shen Zhou and Lee-Ping Wang. Unraveling the structural and chemical features of biological short hydrogen bonds. Chemical Science, 10(32):7734–7745, 2019

    2019

  22. [30]

    Mutation informatics: Sars-cov-2 receptor-binding domain of the spike protein

    Shailendra Verma, Vishal M Patil, and Manish Kumar Gupta. Mutation informatics: Sars-cov-2 receptor-binding domain of the spike protein. Drug Discovery Today, 27(4):103312, 2022

    2022

  23. [31]

    Local weighting schemes for protein multiple sequence alignment

    Jaap Heringa. Local weighting schemes for protein multiple sequence alignment. Computational Chemistry, 26(5):459–477, 2002

    2002

  24. [32]

    Array programming with numpy

    Charles R Harris et al. Array programming with numpy. Nature, 585(7825):357–362, 2020

    2020

  25. [33]

    pandas-dev/pandas: Pandas 1.0.3

    The pandas development team. pandas-dev/pandas: Pandas 1.0.3. Zenodo, 2020

    2020

  26. [34]

    Low: Training deep neural networks by learning optimal sample weights

    Cláudio Santiago, Catarina Barata, Michele Sasdelli, Gustavo Carneiro, and Jacinto C Nascimento. Low: Training deep neural networks by learning optimal sample weights. Pattern Recognition, 107:107585, 2020

    2020

  27. [35]

    Tensorflow: Large-scale machine learning on heterogeneous systems

    Martín Abadi et al. Tensorflow: Large-scale machine learning on heterogeneous systems. 2015

    2015

  28. [36]

    GitHub - keras-team/keras: Deep learning for humans

  29. [37]

    Optuna: A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2623–2631, 2019

    2019

  30. [38]

    Deep learning using rectified linear units (relu)

    Abien Fred Agarap. Deep learning using rectified linear units (relu). arXiv preprint arXiv:1803.08375, 2018

    2018 arXiv

  31. [39]

    Smote: Synthetic minority over-sampling technique

    Nitesh V Chawla, Kevin W Bowyer, Lawrence O Hall, and W Philip Kegelmeyer. Smote: Synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16:321–357, 2002

    2002

  32. [40]

    Feature selection, l1 vs

    Andrew Y Ng. Feature selection, l1 vs. l2 regularization, and rotational invariance. In Proceedings of the Twenty-first International Conference on Machine Learning, page 78, 2004

    2004

  33. [41]

    Probabilistic extension of precision, recall, and f1 score for more thorough evaluation of classification models

    Reb Yacouby and Daniel Axman. Probabilistic extension of precision, recall, and f1 score for more thorough evaluation of classification models. In Proceedings of the First Workshop on Evaluation and Comparison of NLP Systems, pages 79–91, 2020

    2020

  34. [42]

    Juan Terven, Diana-Margarita Cordova-Esparza, Julio-Alejandro Romero-González, Alfonso Ramírez-Pedraza, and E. A. Chávez-Urbiola. A comprehensive survey of loss functions and metrics in deep learning. Artificial Intelligence Review, 58(7):195, apr 2025

    2025

  35. [43]

    Predicting covid-19 disease severity from sars-cov-2 spike protein sequence by mixed effects machine learning

    Bahrad A Sokhansanj and Gail L Rosen. Predicting covid-19 disease severity from sars-cov-2 spike protein sequence by mixed effects machine learning. Computers in Biology and Medicine, 149:105969, 2022

    2022

  36. [44]

    Interpretable and predictive deep neural network modeling of the sars-cov-2 spike protein sequence to predict covid-19 disease severity

    Bahrad A Sokhansanj, Zheng Zhao, and Gail L Rosen. Interpretable and predictive deep neural network modeling of the sars-cov-2 spike protein sequence to predict covid-19 disease severity. Biology, 11(5), 2022

    2022

  37. [45]

    Predicting the sequence specificities of dna-and rna-binding proteins by deep learning

    Babak Alipanahi, Andrew Delong, Matthew T Weirauch, and Brendan J Frey. Predicting the sequence specificities of dna-and rna-binding proteins by deep learning. Nature Biotechnology, 33(8):831–838, 2015

    2015

  38. [46]

    Convolutional neural network architectures for predicting dna-protein binding

    Haoyang Zeng, Matthew D Edwards, Ge Liu, and David K Gifford. Convolutional neural network architectures for predicting dna-protein binding. Bioinformatics, 32(12):i121–i127, 2016

    2016

  39. [47]

    Deep generative models of genetic variation capture the effects of mutations

    Adam J Riesselman, John B Ingraham, and Debora S Marks. Deep generative models of genetic variation capture the effects of mutations. Nature Methods, 15(10):816–822, 2018

    2018

  40. [48]

    Efficient and targeted covid-19 border testing via reinforcement learning

    Hamsa Bastani et al. Efficient and targeted covid-19 border testing via reinforcement learning. Nature, 599(7883):108–113, 2021

    2021

  41. [49]

    Deep learning regularization techniques to genomics data

    Hamza Soumare, Alia Benkahla, and Narjes Gmati. Deep learning regularization techniques to genomics data. Array, 11:100068, 2021. 11 CNN-lSTM Hybrid Model For AI-Driven Prediction Of COVID-19 Severity From Spike Sequences And Clinical Data

    2021

  42. [50]

    Bias in error estimation when using cross-validation for model selection

    Sudhir Varma and Richard Simon. Bias in error estimation when using cross-validation for model selection. BMC Bioinformatics, 7(1):91, 2006

    2006

  43. [51]

    Early detection of sars-cov-2 p.1 variant in southern brazil and reinfection of the same patient by p.2

    Marilda S da Silva et al. Early detection of sars-cov-2 p.1 variant in southern brazil and reinfection of the same patient by p.2. Revista do Instituto de Medicina Tropical de São Paulo, 63:e58, 2021

    2021

  44. [52]

    Genomics and epidemiology of the p.1 sars-cov-2 lineage in manaus, brazil

    Nuno R Faria et al. Genomics and epidemiology of the p.1 sars-cov-2 lineage in manaus, brazil. Science, 372(6544):815–821, 2021

    2021

  45. [53]

    Predominance of the sars-cov-2 lineage p.1 and its sublineage p.1.2 in patients from the metropolitan region of porto alegre, southern brazil in march 2021

    Vitor B Franceschi et al. Predominance of the sars-cov-2 lineage p.1 and its sublineage p.1.2 in patients from the metropolitan region of porto alegre, southern brazil in march 2021. Pathogens, 10(7), 2021

    2021

  46. [54]

    Genomic surveillance and sequencing of SARS-CoV-2 across south america

    Maria Fernanda Ribeiro Dias, Bruce Veiga Andriolo, Diego Henrique Silvestre, Paula Lopes Cascabulho, and Manuela Leal da Silva. Genomic surveillance and sequencing of SARS-CoV-2 across south america. Revista panamericana de salud publica = Pan American journal of public health...

    2023

  47. [55]

    Sars-cov-2 mutations and covid-19 clinical outcome: Mutation global frequency dynamics and structural modulation hold the key

    Ravi Maurya et al. Sars-cov-2 mutations and covid-19 clinical outcome: Mutation global frequency dynamics and structural modulation hold the key. Frontiers in Cellular and Infection Microbiology, 12:868414, 2022

    2022

  48. [56]

    Sars-cov-2 variants and clinical outcomes: A systematic review

    Ivan R Mendiola-Pastrana et al. Sars-cov-2 variants and clinical outcomes: A systematic review. Life, 12(1), 2022

    2022

  49. [57]

    A comprehensive analysis of structural and functional changes induced by sars-cov-2 spike protein mutations

    Aganze Gloire-Aimé Mushebenge, Samuel C Ugbaja, Nonkululeko A Mbatha, Rene M Khan, and Hezekiel M Kumalo. A comprehensive analysis of structural and functional changes induced by sars-cov-2 spike protein mutations. Preprints, 2023

    2023

  50. [58]

    Sars-cov-2 variants, spike mutations and immune escape

    William T Harvey et al. Sars-cov-2 variants, spike mutations and immune escape. Nature Reviews Microbiology, 19(7):409–424, 2021

    2021

  51. [59]

    Reduced sensitivity of sars-cov-2 variant delta to antibody neutralization

    Delphine Planas et al. Reduced sensitivity of sars-cov-2 variant delta to antibody neutralization. Nature, 596(7871):276–280, 2021

    2021

  52. [60]

    Evidence of escape of sars-cov-2 variant b.1.351 from natural and vaccine-induced sera

    Daming Zhou et al. Evidence of escape of sars-cov-2 variant b.1.351 from natural and vaccine-induced sera. Cell, 184(9):2348–2361, 2021

    2021

  53. [61]

    Research on expansion and classification of imbalanced data based on smote algorithm

    Shuo Wang, Yiqin Dai, Jun Shen, and Jifeng Xuan. Research on expansion and classification of imbalanced data based on smote algorithm. Scientific Reports, 11(1):24039, 2021

    2021

  54. [62]

    Challenges and limitations of synthetic minority oversampling techniques in machine learning

    Ibrahim M Alkhawaldeh, Ibrahem Albalkhi, and Ahmad J Naswhan. Challenges and limitations of synthetic minority oversampling techniques in machine learning. World Journal of Methodology, 13(5):373–378, 2023

    2023

  55. [63]

    Unified rational protein engineering with sequence-based deep representation learning

    Ethan C Alley, Grigory Khimulya, Surojit Biswas, Mohammed AlQuraishi, and George M Church. Unified rational protein engineering with sequence-based deep representation learning. Nature Methods, 16(12):1315–1322, 2019. 12

    2019

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.