Genomic prediction using machine learning: a comparison of the performance of regularized regression, ensemble, instance-based and deep learning methods on synthetic and empirical data.

BACKGROUND: The accurate prediction of genomic breeding values is central to genomic selection in both plant and animal breeding studies. Genomic prediction involves the use of thousands of molecular markers spanning the entire genome and therefore requires methods able to efficiently handle high dimensional data. Not surprisingly, machine learning methods are becoming widely advocated for and used in genomic prediction studies. These methods encompass different groups of supervised and unsupervised learning methods. Although several studies have compared the predictive performances of individual methods, studies comparing the predictive performance of different groups of methods are rare. However, such studies are crucial for identifying (i) groups of methods with superior genomic predictive performance and assessing (ii) the merits and demerits of such groups of methods relative to each other and to the established classical methods. Here, we comparatively evaluate the genomic predictive performance and informally assess the computational cost of several groups of supervised machine learning methods, specifically, regularized regression methods, deep, ensemble and instance-based learning algorithms, using one simulated animal breeding dataset and three empirical maize breeding datasets obtained from a commercial breeding program. RESULTS: Our results show that the relative predictive performance and computational expense of the groups of machine learning methods depend upon both the data and target traits and that for classical regularized methods, increasing model complexity can incur huge computational costs but does not necessarily always improve predictive accuracy. Thus, despite their greater complexity and computational burden, neither the adaptive nor the group regularized methods clearly improved upon the results of their simple regularized counterparts. This rules out selection of one procedure among machine learning methods for routine use in genomic prediction. The results also show that, because of their competitive predictive performance, computational efficiency, simplicity and therefore relatively few tuning parameters, the classical linear mixed model and regularized regression methods are likely to remain strong contenders for genomic prediction. CONCLUSIONS: The dependence of predictive performance and computational burden on target datasets and traits call for increasing investments in enhancing the computational efficiency of machine learning algorithms and computing resources.

CT- versus MRI-Based Imaging for Thrombolysis and Mechanical Thrombectomy in Ischemic Stroke: Analysis from the Austrian Stroke Registry.

BACKGROUND AND PURPOSE: It is unclear whether a particular stroke imaging modality offers an advantage for the acute stroke treatment. The aim of this study was to compare procedure times, efficacy and safety of thrombolysis and/or thrombectomy based on computed tomography (CT) versus magnetic resonance imaging (MRI) acute stroke imaging. METHODS: Data of stroke patients who received intravenous thrombolysis (IVT) and/or mechanical thrombectomy (MT) were extracted from a nationwide, prospective stroke unit registry and categorized according to initial imaging modality. Study endpoints included procedure times, symptomatic intracerebral hemorrhage (sICH), early neurological improvement, 3-month functional outcome by modified Rankin Scale (mRS) and mortality. RESULTS: Stroke patients (n=16,799) treated with IVT and 2,248 treated with MT were included. MRI-guided patients (n=2,599) were younger, had less comorbidities and higher rates of strokes with unknown onset as compared to CT-guided patients. In patients treated with IVT, no differences were observed regarding the rates of functional outcome by mRS 0-1 (adjusted odds ratio [OR], 0.87; 95% confidence interval [CI], 0.71 to 1.05), sICH (adjusted OR, 0.82; 95% CI, 0.61 to 1.08), and mortality (adjusted OR, 0.88; 95% CI, 0.63 to 1.22). Patients undergoing MT selected by MRI as compared to CT showed equal rates of functional outcome by mRS 0-2 (adjusted OR, 0.87; 95% CI, 0.65 to 1.16), sICH (adjusted OR, 0.9; 95% CI, 0.51 to 1.69), and mortality (adjusted OR, 0.62; 95% CI, 0.35 to 1.09). MRI-guided patients showed a significant intrahospital delay of about 20 minutes in both the IVT and the MT group. CONCLUSIONS: This large non-randomized comparison study indicates that CT- and MRI-guided patient selection for IVT/MT may perform equally well in terms of functional outcome and safety.

Mechanical Thrombectomy in Acute Stroke Patients with Moderate to Severe Pre-Stroke Disability.

BACKGROUND AND PURPOSE: Studies on mechanical thrombectomy (MT) in acute ischemic stroke (AIS) patients with preexisting disability are limited. We aimed to compare the outcomes of MT versus best medical treatment (BMT) in these patients. METHODS: In the nationwide Austrian registry and Swiss monocentric registry, we identified 462 AIS patients with pre-stroke disability (modified Rankin Scale [mRS] score ≥3) and acute large vessel occlusion. The primary outcome was returning to pre-stroke mRS or better at 3 months. Secondary outcomes were early neurological improvement (National Institutes of Health Stroke Scale score improvement ≥8 at 24 to 48 hours), 3-month mortality, and symptomatic intracerebral hemorrhage (sICH). Multivariable regression models and propensity score matching (PSM) were used for statistical analyses. RESULTS: Compared with the BMT group (n=175), the MT group (n=175) had younger age, more severe strokes, and lower pre-stroke mRS, but similar proportion of receiving intravenous thrombolysis. MT was associated with higher odds of returning to baseline mRS or better at 3 months (adjusted odds ratio [aOR], 2.5; 95% confidence interval [CI], 1.39 to 4.47), early neurological improvement (aOR, 2.62; 95% CI, 1.41 to 4.88), and lower risk of 3-month mortality (aOR, 0.29; 95% CI, 0.18 to 0.49). PSM analysis showed similar findings. MT was not associated with an increased risk of sICH (4.0% vs. 2.1% in all patients; 4.2% vs. 2.4% in the PSM cohort). CONCLUSIONS: MT in patients with pre-stroke mRS ≥3 might improve the 3-month outcomes and short-term neurological impairment, suggesting that pre-stroke disability alone should not be a reason to withhold MT, but that individual case-by-case decisions may be more appropriate.

Time trends in stroke severity in the years 2005 to 2020: results from the Austrian Stroke Unit Registry.

BACKGROUND AND PURPOSE: With aging population, there is an increase of atrial fibrillation (AF) and other vascular risk factors. We investigated trends in stroke severity at hospital admission with respect to AF and other risk factors in a prospective national stroke registry from 2005 to 2020. METHODS: Data from the prospective Austrian Stroke Unit Registry were used to study demographic and clinical factors associated with the change in admission stroke severity over years. Time trends in admission stroke severity of patients with pre-stroke modified Rankin Score ≤ 3 were investigated with respect to clinical variables and predefined age groups 18-54, 55-64, 65-74, 75-84 and ≥ 85 years. Time trends were studied using robust generalized linear models assuming normal distribution with a log link. Stroke severity on admission was assessed according to the National Institutes of Health Stroke Scale Score (NIHSS). RESULTS: In total, 140,312 patients with acute ischemic stroke were included in the analysis. Within the study period, mean patients' age increased from 70 to 72 years (p < 0.001) and median NIHSS at admission decreased from 4 to 3 (p < 0.001). The frequency of AF increased from 25 to 32% (p < 0.001). The decrease in median admission NIHSS was evident in all relevant subgroups but more pronounced in patients with risk factors including AF, diabetes, hypercholesterolemia, smoking, elderly patients and those with pre-stroke disability. CONCLUSION: Despite an aging population and generally increasing AF frequency, we observed a consistent trend towards less disabling strokes on admission.

Thrombectomy in basilar artery occlusion.

BACKGROUND AND PURPOSE: The benefit of thrombectomy (TE) for acute ischemic stroke (AIS) in patients suffering basilar artery occlusion (BAO) is still unclear. Our aim was to analyze functional outcome after 3 months in BAO compared to anterior circulation large vessel occlusion (ACLVO) in a nationwide registry. METHODS: Patients enrolled into the Austrian Endostroke Registry from 2013 to 2018 were analyzed. We used propensity score matching to control for imbalances and to compare patients with BAO and ACLVO. The primary outcome was favorable functional outcome after 3 months measured by the modified Rankin Scale (mRS) (0-2). Multivariate models were applied to estimate the effect of localization (BAO vs ACLVO). RESULTS: In total, 2288 patients underwent TE for AIS with proximal vessel occlusion, of these 267 with BAO. Two hundred and sixty-four patients with BAO were matched to 264 patients with ACLVO. Baseline characteristics were well-balanced. The 90-day mortality did not significantly differ between patients with BAO and ACLVO. In a multivariate logistic regression model, we did not detect a significant difference in functional outcome between BAO and ACLVO (odds ratio for favorable outcome defined as mRS = 0-2: 1.19; 95% confidence interval (CI) = 0.78-1.81; p = 0.42). In patients with an onset-to-door-time ⩾270 min, TE of BAO was associated with poor functional outcome defined as mRS 3-6 (odds ratio (OR) = 3.97; 95% CI = 1.32-11.94; p = 0.01) as compared to ACLVO. CONCLUSION: In this study, functional outcome did not differ after TE in patients with BAO and ACLVO overall; however, we detected an association of BAO with poor outcome in patients arriving late.

Intravenous thrombolysis in stroke with admission NIHSS score 0 or 1.

BACKGROUND: Up to 30% of stroke patients initially presenting with non-disabling or mild deficits may experience poor functional outcome. Despite, intravenous thrombolysis remains controversial in this subgroup of stroke patients due to its uncertain risk benefit ratio. AIM: We aimed to analyze the real-world experience with intravenous thrombolysis in stroke patients presenting with very low NIHSS. METHODS: Data of stroke patients presenting with mild initial stroke severity (NIHSS 0-5) including vascular risk factors, stroke syndrome and etiology, early neurological deterioration, symptomatic intracerebral haemorrhage (sICH), and functional outcome by modified Rankin Scale were extracted from a large nationwide stroke registry and analysed. Patients were categorized and compared according to admission severity NIHSS 0-1 versus NIHSS 2-5 and intravenous thrombolysis use. RESULTS: Seven hundred and three (2%) of 35,113 patients presenting with NIHSS 0-1 and 6316 (13.9%) of 45,521 of patients presenting with NIHSS 2-5 underwent intravenous thrombolysis. In the NIHSS 0-1 group, intravenous thrombolysis was associated with early neurological deterioration (adjusted OR 8.84, CI 6.61-11.83), sICH (adjusted OR 9.32, CI 4.53-19.15) and lower rate of excellent outcome (mRS 0-1) at three months (adjusted OR 0.67, CI 0.5-0.9). In stroke patients with NIHSS 2-5, intravenous thrombolysis was associated with early neurological deterioration (adjusted OR 1.7, 1.47-1.98), sICH (adjusted OR 5.75, CI 4.45-7.45), and higher rate of excellent outcome (mRS 0-1) at three months (adjusted OR 1.21, CI 1.08-1.34). CONCLUSIONS: Among patients with NIHSS 0-1, intravenous thrombolysis did not increase the likelihood of excellent outcome. Moreover, potential signals of harm were observed. Further research seems to be warranted.

Correction to: Integrative transcriptomics reveals genotypic impact on sugar beet storability.

In the above mentioned publication, part of Fig. 6B was distorted (extra diagonal lines appeared). The original article has been corrected and the proper version of Fig. 6B is also published here.

Integrative transcriptomics reveals genotypic impact on sugar beet storability.

KEY MESSAGE: An integrative comparative transcriptomic approach on six sugar beet varieties showing different amount of sucrose loss during storage revealed genotype-specific main driver genes and pathways characterizing storability. Sugar beet is next to sugar cane one of the most important sugar crops accounting for about 15% of the sucrose produced worldwide. Since its processing is increasingly centralized, storage of beet roots over an extended time has become necessary. Sucrose loss during storage is a major concern for the sugar industry because the accumulation of invert sugar and byproducts severely affect sucrose manufacturing. This loss is mainly due to ongoing respiration, but changes in cell wall composition and pathogen infestation also contribute. While some varieties can cope better during storage, the underlying molecular mechanisms are currently undiscovered. We applied integrative transcriptomics on six varieties exhibiting different levels of sucrose loss during storage. Already prior to storage, well storable varieties were characterized by a higher number of parenchyma cells, a smaller cell area, and a thinner periderm. Supporting these findings, transcriptomics identified changes in genes involved in cell wall modifications. After 13 weeks of storage, over 900 differentially expressed genes were detected between well and badly storable varieties, mainly in the category of defense response but also in carbohydrate metabolism and the phenylpropanoid pathway. These findings were confirmed by gene co-expression network analysis where hub genes were identified as main drivers of invert sugar accumulation and sucrose loss. Our data provide insight into transcriptional changes in sugar beet roots during storage resulting in the characterization of key pathways and hub genes that might be further used as markers to improve pathogen resistance and storage properties.

Trends of r-tPA (Recombinant Tissue-Type Plasminogen Activator) Treatment and Treatment-Influencing Factors in Acute Ischemic Stroke.

Background and Purpose- Frequencies of treatment with r-tPA (recombinant tissue-type plasminogen activator) are increasing over the past 15 years. However, published data on the influence of various demographic and clinical factors on r-tPA treatment as well as estimates of future trajectories are limited. We evaluated time trends and future trajectories of r-tPA treatment in patients with acute stroke and the influence of various factors on r-tPA treatment by analyzing data of 103 970 patients enrolled in the Austrian Stroke Unit Registry from 2006 to 2018, of which 18 953 were treated with r-tPA. Methods- Time trends of r-tPA-treatment were investigated in predefined subgroups (minor/major stroke, age, anterior/posterior circulation stroke); limited exponential time series models were calculated to estimate future trends of r-tPA-treatment. Logistic regression models were calculated to estimate the influence of clinical variables on r-tPA-treatment. Results- Overall, r-tPA treatment frequencies increased from 9.9% in 2006 to 21.8% in 2018. We observed a particular increase in patients >80 years, patients presenting with a National Institutes of Health Stroke Scale Score of 2 to 3, patients with posterior circulation stroke, patients with wake-up stroke, and patients without atrial fibrillation. Forecast of overall r-tPA frequencies predicted a further but flattened increase up to 24% by 2025. Logistic regression of time-dependent associations of clinical variables with r-tPA-treatment revealed increasing odds of r-tPA-treatment in patients with a posterior circulation stroke and decreasing odds of r-tPA-treatment in patients with atrial fibrillation. Conclusions- We observed a positive development of r-tPA-treatment frequencies mirroring increasing confidence with intravenous thrombolysis in clinical practice; however, decreasing odds of r-tPA-treatment over time in patients with atrial fibrillation deserve particular attention.

Temporal trends in intracerebral hemorrhage: Evidence from the Austrian Stroke Unit Registry.

BACKGROUND: To assess changes in frequency, severity, complications, therapy and outcome of intracerebral hemorrhage in patients treated in stroke units in Austria, we evaluated data from the Austrian Stroke Unit Registry between 2008 and 2016. METHODS AND FINDINGS: Data of 6707 cases of ICH covering a time span of 9 years and including information on age, risk factors, pre-stroke modified Rankin Score (mRS), baseline stroke severity (NIHSS), complications, therapy, functional outcome, and mortality were extracted from the Austrian Stroke Unit Registry. A multivariate regularized logistic regression model and linear models for temporal dependence were computed for analyzing statistical inference and time trends. Bonferroni correction was applied to correct for multiple testing. Between 2008 and 2016, the proportion of ICH admissions to stroke units in Austria declined, with a shift among patients towards older age (>70 years, p = 0.04) and lower admission NIHSS scores. While no significant time trends in risk factors, pre-stroke mRS and medical complications were observed, therapeutic interventions declined significantly (p<0.001). Three-month mortality increased over the years independently (p = 0.003). CONCLUSIONS: Despite declining incidence and clinical severity of ICH we observed a clear increase in three-month mortality. This effect seems to be independent of predictors including age, admission NIHSS, pre-morbid MRS, or medical complications. The observations from this large retrospective database cohort study underline an urgent call for action in the therapy of ICH.

Stroke risk factors and treatment variables in rural and urban Austria: An analysis of the Austrian Stroke Unit Registry.

BACKGROUND AND OBJECTIVES: Differences in stroke risk factors and treatment variables between rural and urban regions in Austria were analyzed retrospectively as European data on this topic are scarce. RESEARCH DESIGN AND METHODS: We performed statistical analysis using group comparisons and time series analysis of data of the Austrian Stroke Unit Registry between 2005 and 2016. 87411 patients were divided into three groups (rural, intermediate, urban) according to the degree of urbanisation classification of the European Commission/Eurostat. RESULTS: Patients in the rural group were significantly younger, more often female, had a lower pre-stroke disability, and were more frequently transported by an emergency physician. Vascular risk factors were significantly higher in urban patients, leading to a higher rate of microangiopathic etiology. Onset-to-door (ODT) and Onset-to-treatment times were significantly higher in the rural group, but ODTs decreased over time. Door-to-needle times and time to first vascular imaging were significantly lower in the rural group. Intravenous thrombolysis and rehabilitation rates were lower in urban patients. DISCUSSION AND IMPLICATIONS: Contrary to previous literature predominantly from outside of Europe, vascular risk factors were higher in Austrian urban patients. Further, rural patients had higher intravenous thrombolysis and rehabilitation rates maybe because of lower pre-stroke disability. ODTs in rural patients were generally higher, but they decreased over time, which might be a consequence of better education of the public in noticing early stroke signs, better transportation and education of emergency medical personnel, better advance notification to the receiving hospital and implementation of Stroke Units in rural areas.

Modified DWI-FLAIR mismatch guided thrombolysis in unknown onset stroke.

DWI-FLAIR mismatch has been recently proven to identify patients with unknown onset stroke (UOS) eligible for thrombolysis. However, this concept may exclude patients from thrombolysis who may eventually benefit as well. We aimed to examine the feasibility, safety and potential efficacy of thrombolysis in wake-up stroke (WUS) and UOS patients using a modified DWI-FLAIR mismatch allowing for partial FLAIR positivity. WUS/UOS patients fulfilling the modified DWI-FLAIR mismatch and treated with intravenous thrombolysis (IVT) were compared to propensity score matched WUS/UOS patients excluded from IVT due to FLAIR positivity. The primary endpoint was a symptomatic intracranial hemorrhage (SICH), the secondary endpoints were improvement of ≥ 4 in NIHSS score and mRS score at 3 months. 64 IVT-treated patients (median NIHSS 9) and 64 controls (median NIHSS 8) entered the analysis (p = 0.2). No significant difference in SICH was found between the IVT group and the controls (3.1% vs. 1.6%, p = 0.9). An improvement of ≥ 4 NIHSS points was more frequent in IVT patients as compared to controls (40.6% vs. 18.8%, p = 0.01). 23.4% of IVT patients achieved a mRS score of 0-1 at 3 months as compared to 18.8% of the controls (p = 0.8). SICH, improvement of NIHSS ≥ 4 and mRS 0-1 at 3 months were comparable in thrombolyzed patients with negative FLAIR images versus those thrombolyzed with partial positive FLAIR images (3% vs. 3%, p = 0.9; 40% vs. 41%, p = 0.9; 19% vs. 22%, p = 0.8). Our study signalizes that thrombolysis may be feasible in selected WUS/UOS patients with partial FLAIR signal positivity.

Predicting Early Mortality of Acute Ischemic Stroke.

Background and Purpose- Several risk factors are known to increase mid- and long-term mortality of ischemic stroke patients. Information on predictors of early stroke mortality is scarce but often requested in clinical practice. We therefore aimed to develop a rapidly applicable tool for predicting early mortality at the stroke unit. Methods- We used data from the nationwide Austrian Stroke Unit Registry and multivariate regularized logistic regression analysis to identify demographic and clinical variables associated with early (≤7 days poststroke) mortality of patients admitted with ischemic stroke. These variables were then used to develop the Predicting Early Mortality of Ischemic Stroke score that was validated both by bootstrapping and temporal validation. Results- In total, 77 653 ischemic stroke patients were included in the analysis (median age: 74 years, 47% women). The mortality rate at the stroke unit was 2% and median stay of deceased patients was 3 days. Age, stroke severity measured by the National Institutes of Health Stroke Scale, prestroke functional disability (modified Rankin Scale >0), preexisting heart disease, diabetes mellitus, posterior circulation stroke syndrome, and nonlacunar stroke cause were associated with mortality and served to build the Predicting Early Mortality of Ischemic Stroke score ranging from 0 to 12 points. The area under the curve of the score was 0.879 (95% CI, 0.871-0.886) in the derivation cohort and 0.884 (95% CI, 0.863-0.905) in the validation sample. Patients with a score ≥10 had a 35% (95% CI, 28%-43%) risk to die within the first days at the stroke unit. Conclusions- We developed a simple score to estimate early mortality of ischemic stroke patients treated at a stroke unit. This score could help clinicians in short-term prognostication for management decisions and counseling.

Is Functional Outcome Different in Posterior and Anterior Circulation Stroke?

Background and Purpose- Posterior circulation stroke (PCS) account for 20% of all ischemic strokes. There is limited evidence whether functional outcome of PCS is comparable to that of anterior circulation stroke (ACS). We aimed to analyze whether 3-month functional outcome is different in PCS and ACS. Methods- Patients with acute ischemic stroke prospectively enrolled within the Austrian Stroke Unit Registry were stratified by infarct localization according to the Oxfordshire Community Stroke Project Classification. Propensity score matching was used to control for covariate imbalances and to match patients with PCS and ACS. Patients were matched for stroke severity, recombinant tissue-type plasminogen activator treatment, and demographic and vascular risk factors. Main outcomes were the distribution of modified Rankin Scale after 3 months and multiple proportional odds models to estimate the influence of the infarct localization on the functional outcome. Results- From a total of 90 484 patients enrolled within the Austrian Stroke Unit Registry, 9208 (4604 PCS/4604 ACS) were matched, of those 954 (477 in each group) were treated with recombinant tissue-type plasminogen activator. We detected a significant shift towards better 3-month functional outcome in patients with ACS compared with PCS (odds ratio [OR], 1.19; 95% CI, 1.1-1.28; P<0.0001). In particular, functional outcome was worse in PCS with onset-to-door-time >270 minutes (OR, 1.34; 95% CI, 1.17-1.54; P<0.0001) and in PCS with unknown onset-to-door-time (OR, 1.26; 95% CI, 1.13-1.42; P<0.0001); however, we did not detect any difference in functional outcome between ACS and PCS in patients with an onset-to-door-time ≤270 minutes (1-180 minutes: OR, 0.92, 95% CI, 0.78-1.09, P=0.3554; 181-270 minutes: OR, 1.04, 95% CI, 0.79-1.37, P=0.7689). In patients treated with recombinant tissue-type plasminogen activator, functional outcome was not significantly different between PCS and ACS. Conclusions- PCS was associated with worse outcome compared with ACS in patients arriving later than 4.5 hours at hospital or in those with unknown onset of symptoms. Our results urge for implementation of symptoms found in the posterior circulation into preclinical patient-triage tools.

Prehospital and intra-hospital time delays in posterior circulation stroke: results from the Austrian Stroke Unit Registry.

Therapeutic effect of recombinant tissue-plasminogen activator (rt-PA) is time dependent. There is limited evidence whether localization of stroke within the posterior circulation (PCS) is associated with a treatment delay. We aimed to analyze within a nationwide multicenter cohort whether duration of pre- and intra-hospital patient management differs between patients with PCS and anterior circulation strokes (ACS). We studied onset-to-door-times (ODT) and door-to-needle-times (DNT) of all patients with acute ischemic stroke (IS) enrolled in the Austrian Stroke Unit Registry according to infarct localization. Classification into PCS and ACS was based on clinical presentation applying the criteria used in the Oxfordshire Community Stroke Project. Relationships between ODT, respectively, DNT and explanatory variables were modeled by multivariate linear regression. Between 2003 and 2015, 71010 patients with IS were enrolled, 11,924 with PCS and 59,086 with ACS. Overall, the ODT was significantly longer in PCS: median (IQR): 170 (25th, 75th‰: 79,420) min versus 110 (60,240); p < 0.001; this finding held true in multivariable analysis. In 10535 rt-PA-treated patients (1022 PCS/9832 ACS), ODT and DNT were significantly longer among those with PCS: ODT: median: 80 min (55,120) versus 72 (50,110), p < 0.001; DNT: 57 (35.90) versus 45 (30.67), p < 0.001. In the multivariate model, PCS was significantly associated with delay in the DNT. In conclusion, in this large nationwide cohort, patient management was significantly slower in PCS as compared to ACS. Increasing awareness about these delays and further elaboration of the underlying causes may translate into higher proportions of patients with PCS receiving rt-PA.

Hybridization thermodynamics of NimbleGen microarrays.

BACKGROUND: While microarrays are the predominant method for gene expression profiling, probe signal variation is still an area of active research. Probe signal is sequence dependent and affected by probe-target binding strength and the competing formation of probe-probe dimers and secondary structures in probes and targets. RESULTS: We demonstrate the benefits of an improved model for microarray hybridization and assess the relative contributions of the probe-target binding strength and the different competing structures. Remarkably, specific and unspecific hybridization were apparently driven by different energetic contributions: For unspecific hybridization, the melting temperature Tm was the best predictor of signal variation. For specific hybridization, however, the effective interaction energy that fully considered competing structures was twice as powerful a predictor of probe signal variation. We show that this was largely due to the effects of secondary structures in the probe and target molecules. The predictive power of the strength of these intramolecular structures was already comparable to that of the melting temperature or the free energy of the probe-target duplex. CONCLUSIONS: This analysis illustrates the importance of considering both the effects of probe-target binding strength and the different competing structures. For specific hybridization, the secondary structures of probe and target molecules turn out to be at least as important as the probe-target binding strength for an understanding of the observed microarray signal intensities. Besides their relevance for the design of new arrays, our results demonstrate the value of improving thermodynamic models for the read-out and interpretation of microarray signals.