A Random Forest Classifier Uses Antibody Responses to Plasmodium Antigens to Reveal Candidate Biomarkers of the Intensity and Timing of Past Exposure to Plasmodium falciparum

Abstract

Important goals of malaria surveillance efforts include accurately quantifying the burden of malaria over time, which can be useful to target and evaluate interventions. The majority of malaria surveillance methods capture active or recent infections which poses several challenges to achieving malaria surveillance goals. In high transmission settings, asymptomatic infections are common and therefore accurate measurement of malaria burden often demands active surveillance; in low transmission regions where infections are rare accurate surveillance requires sampling a large subset of the population; and in any context monitoring malaria burden over time necessitates serial sampling. Antibody responses to Plasmodium falciparum parasites persist after infection and therefore measuring antibodies has the potential to overcome several of the current difficulties associated with malaria surveillance. However, identifying which antibody responses are markers of the timing and intensity of past exposure to P. falciparum is challenging, particularly among adults who tend to be re-exposed multiple times over the course of their lifetime and therefore have similarly high antibody responses to many P. falciparum antigens. A previous analysis of 479 serum samples from individuals in three regions in southern Africa with different historical levels of P. falciparum malaria transmission (high, intermediate, and low) revealed regional differences in antibody responses to P. falciparum antigens among children under 5 years of age. Using a novel bioinformatic pipeline optimized for protein microarrays that minimizes between-sample technical variation, we used antibody responses to P. falciparum and P. vivax antigens as predictors in random forest models to classify adult samples into these three regions of differing historical malaria transmission with high accuracy. Many of the antigens that were most important for classification in these models do not overlap with previously published results and are therefore novel candidate markers for the timing and intensity of past exposure to P. falciparum. Measuring antibody responses to these antigens could potentially lead to improved malaria serosurveillance that captures the timing and intensity of past exposure.

A Bayesian hierarchical model for signal extraction from protein microarrays

“…Further investigation of the squared error loss ranks for antigens measured by the malaria arrays used in Kobayashi et al 25 revealed potential novel biomarkers of the timing and intensity of past malaria exposure in adult populations. 31 In spite of the differences in manufacturing and features of the arrays used in the two datasets, our model fits quite well to both, suggesting the potential for it to be effective across multiple protein microarray studies. However, the two published studies represent a fairly narrow sample of the currently available protein microarrays, and the amount of between-array variability present is dependent, at least in part, on the manufacturing process.…”

“…Finally, we showed how the ranks of proteins and therefore subsequent analysis can be impacted by the choices of both preprocessing and ranking methods. Further investigation of the squared error loss ranks for antigens measured by the malaria arrays used in Kobayashi et al 25 revealed potential novel biomarkers of the timing and intensity of past malaria exposure in adult populations 31 …”

A Bayesian Hierarchical Model for Signal Extraction from Protein Microarrays

“…In both cases, however, the use of the full posterior distributions to produce optimal ranks under different loss functions substantially changes the ordering and composition of the top 30 proteins list. The statistical literature shows that ranking methods accounting for uncertainty by optimizing a loss function outperform other methods of ranking, (Shen and Louis, 1998; Lin and others , 2006; Henderson and Newton, 2016) therefore ranks for the malaria and lung cancer studies obtained by using the Bayseian model outputs are preferable to those obtained using other estimates of normalized fluorescent intensity (Bérubé and others , 2022).…”