Predicting haemoglobin deferral using machine learning models: Can we use the same prediction model across countries?

BACKGROUND AND OBJECTIVES: Personalized donation strategies based on haemoglobin (Hb) prediction models may reduce Hb deferrals and hence costs of donation, meanwhile improving commitment of donors. We previously found that prediction models perform better in validation data with a high Hb deferral rate. We therefore investigate how Hb deferral prediction models perform when exchanged with other blood establishments. MATERIALS AND METHODS: Donation data from the past 5 years from random samples of 10,000 donors from Australia, Belgium, Finland, the Netherlands and South Africa were used to fit random forest models for Hb deferral prediction. Trained models were exchanged between blood establishments. Model performance was evaluated using the area under the precision-recall curve (AUPR). Variable importance was assessed using SHapley Additive exPlanations (SHAP) values. RESULTS: Across the validation datasets and exchanged models, the AUPR ranged from 0.05 to 0.43. Exchanged models performed similarly within validation datasets, irrespective of the origin of the training data. Apart from subtle differences, the importance of most predictor variables was similar in all trained models. CONCLUSION: Our results suggest that Hb deferral prediction models trained in different blood establishments perform similarly within different validation datasets, regardless of the deferral rate of their training data. Models learn similar associations in different blood establishments.

The value of genetic data from 665,460 individuals in managing iron deficiency anaemia and suitability to donate blood.

BACKGROUND AND OBJECTIVES: Although the genetic determinants of haemoglobin and ferritin have been widely studied, those of the clinically and globally relevant iron deficiency anaemia (IDA) and deferral due to hypohaemoglobinemia (Hb-deferral) are unclear. In this investigation, we aimed to quantify the value of genetic information in predicting IDA and Hb-deferral. MATERIALS AND METHODS: We analysed genetic data from up to 665,460 participants of the FinnGen, Blood Service Biobank and UK Biobank, and used INTERVAL (N = 39,979) for validation. We performed genome-wide association studies (GWASs) of IDA and Hb-deferral and utilized publicly available genetic associations to compute polygenic scores for IDA, ferritin and Hb. We fitted models to estimate the effect sizes of these polygenic risk scores (PRSs) on IDA and Hb-deferral risk while accounting for the individual's age, sex, weight, height, smoking status and blood donation history. RESULTS: Significant variants in GWASs of IDA and Hb-deferral appear to be a small subset of variants associated with ferritin and Hb. Effect sizes of genetic predictors of IDA and Hb-deferral are similar to those of age and weight which are typically used in blood donor management. A total genetic score for Hb-deferral was estimated for each individual. The odds ratio estimate between first decile against that at ninth decile of total genetic score distribution ranged from 1.4 to 2.2. CONCLUSION: The value of genetic data in predicting IDA or suitability to donate blood appears to be on a practically useful level.

The added value of ferritin levels and genetic markers for the prediction of haemoglobin deferral.

BACKGROUND AND OBJECTIVES: On-site haemoglobin deferral for blood donors is sometimes necessary for donor health but demotivating for donors and inefficient for the blood bank. Deferral rates could be reduced by accurately predicting donors' haemoglobin status before they visit the blood bank. Although such predictive models have been published, there is ample room for improvement in predictive performance. We aim to assess the added value of ferritin levels or genetic markers as predictor variables in haemoglobin deferral prediction models. MATERIALS AND METHODS: Support vector machines with and without this information (the full and reduced model, respectively) are compared in Finland and the Netherlands. Genetic markers are available in the Finnish data and ferritin levels in the Dutch data. RESULTS: Although there is a clear association between haemoglobin deferral and both ferritin levels and several genetic markers, predictive performance increases only marginally with their inclusion as predictors. The recall of deferrals increases from 68.6% to 69.9% with genetic markers and from 79.7% to 80.0% with ferritin levels included. Subgroup analyses show that the added value of these predictors is higher in specific subgroups, for example, for donors with minor alleles on single-nucleotide polymorphism 17:58358769, recall of deferral increases from 73.3% to 93.3%. CONCLUSION: Including ferritin levels or genetic markers in haemoglobin deferral prediction models improves predictive performance. The increase in overall performance is small but may be substantial for specific subgroups. We recommend including this information as predictor variables when available, but not to collect it for this purpose only.

An international comparison of haemoglobin deferral prediction models for blood banking.

BACKGROUND AND OBJECTIVES: Blood banks use a haemoglobin (Hb) threshold before blood donation to minimize donors' risk of anaemia. Hb prediction models may guide decisions on which donors to invite, and should ideally also be generally applicable, thus in different countries and settings. In this paper, we compare the outcome of various prediction models in different settings and highlight differences and similarities. MATERIALS AND METHODS: Donation data of repeat donors from the past 5 years of Australia, Belgium, Finland, the Netherlands and South Africa were used to fit five identical prediction models: logistic regression, random forest, support vector machine, linear mixed model and dynamic linear mixed model. Only donors with five or more donation attempts were included to ensure having informative data from all donors. Analyses were performed for men and women separately and outcomes compared. RESULTS: Within countries and overall, different models perform similarly well. However, there are substantial differences in model performance between countries, and there is a positive association between the deferral rate in a country and the ability to predict donor deferral. Nonetheless, the importance of predictor variables across countries is similar and is highest for the previous Hb level. CONCLUSION: The limited impact of model architecture and country indicates that all models show similar relationships between the predictor variables and donor deferral. Donor deferral is found to be better predictable in countries with high deferral rates. Therefore, such countries may benefit more from deferral prediction models than those with low deferral rates.

Prediction and impact of personalized donation intervals.

BACKGROUND AND OBJECTIVES: Deferral of blood donors due to low haemoglobin (Hb) is demotivating to donors, can be a sign for developing anaemia and incurs costs for blood establishments. The prediction of Hb deferral has been shown to be feasible in a number of studies based on demographic, Hb measurement and donation history data. The aim of this paper is to evaluate how state-of-the-art computational prediction tools can facilitate nationwide personalized donation intervals. MATERIALS AND METHODS: Using donation history data from the last 20 years in Finland, FinDonor blood donor cohort data and blood service Biobank genotyping data, we built linear and non-linear predictors of Hb deferral. Based on financial data from the Finnish Red Cross Blood Service, we then estimated the economic impacts of deploying such predictors. RESULTS: We discovered that while linear predictors generally predict Hb relatively well, they have difficulties in predicting low Hb values. Overall, we found that non-linear or linear predictors with or without genetic data performed only slightly better than a simple cutoff based on previous Hb. However, if any of our deferral prediction methods are used to assign temporary prolongations of donation intervals for females, then our calculations indicate cost savings while maintaining the blood supply. CONCLUSION: We find that even though the prediction accuracy is not very high, the actual use of any of our predictors in blood collection is still likely to bring benefits to blood donors and blood establishments alike.

MODER2: first-order Markov modeling and discovery of monomeric and dimeric binding motifs.

MOTIVATION: Position-specific probability matrices (PPMs, also called position-specific weight matrices) have been the dominating model for transcription factor (TF)-binding motifs in DNA. There is, however, increasing recent evidence of better performance of higher order models such as Markov models of order one, also called adjacent dinucleotide matrices (ADMs). ADMs can model dependencies between adjacent nucleotides, unlike PPMs. A modeling technique and software tool that would estimate such models simultaneously both for monomers and their dimers have been missing. RESULTS: We present an ADM-based mixture model for monomeric and dimeric TF-binding motifs and an expectation maximization algorithm MODER2 for learning such models from training data and seeds. The model is a mixture that includes monomers and dimers, built from the monomers, with a description of the dimeric structure (spacing, orientation). The technique is modular, meaning that the co-operative effect of dimerization is made explicit by evaluating the difference between expected and observed models. The model is validated using HT-SELEX and generated datasets, and by comparing to some earlier PPM and ADM techniques. The ADM models explain data slightly better than PPM models for 314 tested TFs (or their DNA-binding domains) from four families (bHLH, bZIP, ETS and Homeodomain), the ADM mixture models by MODER2 being the best on average. AVAILABILITY AND IMPLEMENTATION: Software implementation is available from https://github.com/jttoivon/moder2. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Modular discovery of monomeric and dimeric transcription factor binding motifs for large data sets.

In some dimeric cases of transcription factor (TF) binding, the specificity of dimeric motifs has been observed to differ notably from what would be expected were the two factors to bind to DNA independently of each other. Current motif discovery methods are unable to learn monomeric and dimeric motifs in modular fashion such that deviations from the expected motif would become explicit and the noise from dimeric occurrences would not corrupt monomeric models. We propose a novel modeling technique and an expectation maximization algorithm, implemented as software tool MODER, for discovering monomeric TF binding motifs and their dimeric combinations. Given training data and seeds for monomeric motifs, the algorithm learns in the same probabilistic framework a mixture model which represents monomeric motifs as standard position-specific probability matrices (PPMs), and dimeric motifs as pairs of monomeric PPMs, with associated orientation and spacing preferences. For dimers the model represents deviations from pure modular model of two independent monomers, thus making co-operative binding effects explicit. MODER can analyze in reasonable time tens of Mbps of training data. We validated the tool on HT-SELEX and ChIP-seq data. Our findings include some TFs whose expected model has palindromic symmetry but the observed model is directional.

Conservation of transcription factor binding specificities across 600 million years of bilateria evolution.

Divergent morphology of species has largely been ascribed to genetic differences in the tissue-specific expression of proteins, which could be achieved by divergence in cis-regulatory elements or by altering the binding specificity of transcription factors (TFs). The relative importance of the latter has been difficult to assess, as previous systematic analyses of TF binding specificity have been performed using different methods in different species. To address this, we determined the binding specificities of 242 Drosophila TFs, and compared them to human and mouse data. This analysis revealed that TF binding specificities are highly conserved between Drosophila and mammals, and that for orthologous TFs, the similarity extends even to the level of very subtle dinucleotide binding preferences. The few human TFs with divergent specificities function in cell types not found in fruit flies, suggesting that evolution of TF specificities contributes to emergence of novel types of differentiated cells.

DNA-binding specificities of human transcription factors.

Although the proteins that read the gene regulatory code, transcription factors (TFs), have been largely identified, it is not well known which sequences TFs can recognize. We have analyzed the sequence-specific binding of human TFs using high-throughput SELEX and ChIP sequencing. A total of 830 binding profiles were obtained, describing 239 distinctly different binding specificities. The models represent the majority of human TFs, approximately doubling the coverage compared to existing systematic studies. Our results reveal additional specificity determinants for a large number of factors for which a partial specificity was known, including a commonly observed A- or T-rich stretch that flanks the core motifs. Global analysis of the data revealed that homodimer orientation and spacing preferences, and base-stacking interactions, have a larger role in TF-DNA binding than previously appreciated. We further describe a binding model incorporating these features that is required to understand binding of TFs to DNA.

Multiplexed massively parallel SELEX for characterization of human transcription factor binding specificities.

The genetic code-the binding specificity of all transfer-RNAs--defines how protein primary structure is determined by DNA sequence. DNA also dictates when and where proteins are expressed, and this information is encoded in a pattern of specific sequence motifs that are recognized by transcription factors. However, the DNA-binding specificity is only known for a small fraction of the approximately 1400 human transcription factors (TFs). We describe here a high-throughput method for analyzing transcription factor binding specificity that is based on systematic evolution of ligands by exponential enrichment (SELEX) and massively parallel sequencing. The method is optimized for analysis of large numbers of TFs in parallel through the use of affinity-tagged proteins, barcoded selection oligonucleotides, and multiplexed sequencing. Data are analyzed by a new bioinformatic platform that uses the hundreds of thousands of sequencing reads obtained to control the quality of the experiments and to generate binding motifs for the TFs. The described technology allows higher throughput and identification of much longer binding profiles than current microarray-based methods. In addition, as our method is based on proteins expressed in mammalian cells, it can also be used to characterize DNA-binding preferences of full-length proteins or proteins requiring post-translational modifications. We validate the method by determining binding specificities of 14 different classes of TFs and by confirming the specificities for NFATC1 and RFX3 using ChIP-seq. Our results reveal unexpected dimeric modes of binding for several factors that were thought to preferentially bind DNA as monomers.