Individual risk prediction: Comparing random forests with Cox proportional-hazards model by a simulation study

With big data becoming widely available in healthcare, machine learning algorithms such as random forest (RF) that ignores time-to-event information and random survival forest (RSF) that handles right-censored data are used for individual risk prediction alternatively to the Cox proportional hazards (Cox-PH) model. We aimed to systematically compare RF and RSF with Cox-PH. RSF with three split criteria [log-rank (RSF-LR), log-rank score (RSF-LRS), maximally selected rank statistics (RSF-MSR)]; RF, Cox-PH, and Cox-PH with splines (Cox-S) were evaluated through a simulation study based on real data. One hundred eighty scenarios were investigated assuming different associations between the predictors and the outcome (linear/linear and interactions/nonlinear/nonlinear and interactions), training sample sizes (500/1000/5000), censoring rates (50%/75%/93%), hazard functions (increasing/decreasing/constant), and number of predictors (seven, 15 including noise variables). Methods' performance was evaluated with time-dependent area under curve and integrated Brier score. In all scenarios, RF had the worst performance. In scenarios with a low number of events (⩽70), Cox-PH was at least noninferior to RSF, whereas under linearity assumption it outperformed RSF. Under the presence of interactions, RSF performed better than Cox-PH as the number of events increased whereas Cox-S reached at least similar performance with RSF under nonlinear effects. RSF-LRS performed slightly worse than RSF-LR and RSF-MSR when including noise variables and interaction effects. When applied to real data, models incorporating survival time performed better. Although RSF algorithms are a promising alternative to conventional Cox-PH as data complexity increases, they require a higher number of events for training. In time-to-event analysis, algorithms that consider survival time should be used.     

Complex Patterns of Chromosome 11 Aberrations in Myeloid Malignancies Target CBL,MLL, DDB1 and LMO2

Exome sequencing of primary tumors identifies complex somatic mutation patterns. Assignment of relevance of individual somatic mutations is difficult and poses the next challenge for interpretation of next generation sequencing data. Here we present an approach how exome sequencing in combination with SNP microarray data may identify targets of chromosomal aberrations in myeloid malignancies. The rationale of this approach is that hotspots of chromosomal aberrations might also harbor point mutations in the target genes of deletions, gains or uniparental disomies (UPDs). Chromosome 11 is a frequent target of lesions in myeloid malignancies. Therefore, we studied chromosome 11 in a total of 813 samples from 773 individual patients with different myeloid malignancies by SNP microarrays and complemented the data with exome sequencing in selected cases exhibiting chromosome 11 defects. We found gains, losses and UPDs of chromosome 11 in 52 of the 813 samples (6.4%). Chromosome 11q UPDs frequently associated with mutations of CBL. In one patient the 11qUPD amplified somatic mutations in both CBL and the DNA repair gene DDB1. A duplication within MLL exon 3 was detected in another patient with 11qUPD. We identified several common deleted regions (CDR) on chromosome 11. One of the CDRs associated with de novo acute myeloid leukemia (P=0.013). One patient with a deletion at the LMO2 locus harbored an additional point mutation on the other allele indicating that LMO2 might be a tumor suppressor frequently targeted by 11p deletions. Our chromosome-centered analysis indicates that chromosome 11 contains a number of tumor suppressor genes and that the role of this chromosome in myeloid malignancies is more complex than previously recognized.     

Identification of a Functional Connectome for Long-Term Fear Memory in Mice

Long-term memories are thought to depend upon the coordinated activation of a broad network of cortical and subcortical brain regions. However, the distributed nature of this representation has made it challenging to define the neural elements of the memory trace, and lesion and electrophysiological approaches provide only a narrow window into what is appreciated a much more global network. Here we used a global mapping approach to identify networks of brain regions activated following recall of long-term fear memories in mice. Analysis of Fos expression across 84 brain regions allowed us to identify regions that were co-active following memory recall. These analyses revealed that the functional organization of long-term fear memories depends on memory age and is altered in mutant mice that exhibit premature forgetting. Most importantly, these analyses indicate that long-term memory recall engages a network that has a distinct thalamic-hippocampal-cortical signature. This network is concurrently integrated and segregated and therefore has small-world properties, and contains hub-like regions in the prefrontal cortex and thalamus that may play privileged roles in memory expression.     

Improving Education in Medical Statistics: Implementing a Blended Learning Model in the Existing Curriculum

Background

Although recent studies report on the benefits of blended learning in improving medical student education, there is still no empirical evidence on the relative effectiveness of blended over traditional learning approaches in medical statistics. We implemented blended along with on-site (i.e. face-to-face) learning to further assess the potential value of web-based learning in medical statistics.

Methods

This was a prospective study conducted with third year medical undergraduate students attending the Faculty of Medicine, University of Belgrade, who passed (440 of 545) the final exam of the obligatory introductory statistics course during 2013–14. Student statistics achievements were stratified based on the two methods of education delivery: blended learning and on-site learning. Blended learning included a combination of face-to-face and distance learning methodologies integrated into a single course.

Results

Mean exam scores for the blended learning student group were higher than for the on-site student group for both final statistics score (89.36±6.60 vs. 86.06±8.48; p = 0.001) and knowledge test score (7.88±1.30 vs. 7.51±1.36; p = 0.023) with a medium effect size. There were no differences in sex or study duration between the groups. Current grade point average (GPA) was higher in the blended group. In a multivariable regression model, current GPA and knowledge test scores were associated with the final statistics score after adjusting for study duration and learning modality (p<0.001).

Conclusion

This study provides empirical evidence to support educator decisions to implement different learning environments for teaching medical statistics to undergraduate medical students. Blended and on-site training formats led to similar knowledge acquisition; however, students with higher GPA preferred the technology assisted learning format. Implementation of blended learning approaches can be considered an attractive, cost-effective, and efficient alternative to traditional classroom training in medical statistics.     

Support for Kurdish language rights in Turkey: the roles of ethnic group,group identifications,contact, and intergroup perceptions

The question of Kurdish language rights has been a central issue in the Turkish–Kurdish conflict. The current study examined endorsement of Kurdish language rights in relation to intergroup factors (i.e. group identifications, cross-group friendships, perceived discrimination, and perceived out-group beliefs about state unity) among self-identified Turkish and Kurdish participants. The results indicate that Turks were much less in favour of these rights than the Kurds. In addition, for the Turks, higher national and ethnic identification were associated with lower support for Kurdish language rights, while cross-group friendship, perceived discrimination of Kurds and the belief that Kurds endorse national unity were associated with more support for rights. For the Kurdish participants, stronger national identification seems to undermine the mobilizing meaning that Kurdish group identification has for language rights support. Furthermore, friendship with Turks can undermine the support for rights because it strengthens national identification and reduces ethnic identification.     

Human Intestinal Allografts Contain Functional Hematopoietic Stem and Progenitor Cells that Are Maintained by a Circulating Pool

1. Download high-res image (242KB)
2. Download full-size image

     

Gas exchange is essential for bioreactor cultivation of tissue engineered cartilage

Tissue engineered cartilage can be grown in vitro if the necessary physical and biochemical factors are present in the tissue culture environment. Cell metabolism and tissue composition were studied for engineered cartilage cultured for 5 weeks using bovine articular chondrocytes, polymer scaffolds (5 mm diameter x 2 mm thick fibrous discs), and rotating bioreactors. Medium pH and concentrations of oxygen, carbon dioxide, glucose, lactate, ammonia, and glycosoaminoglycan (GAG) were varied by altering the exchange rates of gas and medium in the bioreactors. Cell-polymer constructs were assessed with respect to histomorphology, biochemical composition and metabolic activity. Low oxygen tension ( approximately 40 mmHg) and low pH ( approximately 6.7) were associated with anaerobic cell metabolism (yield of lactate on glucose, YL/G, of 2.2 mol/mol) while higher oxygen tension ( approximately 80 mmHg) and higher pH ( approximately 7.0) were associated with more aerobic cell metabolism (YL/G of 1.65-1.79 mol/mol). Under conditions of infrequent medium replacement (50% once per week), cells utilized more economical pathways such that glucose consumption and lactate production both decreased, cell metabolism remained relatively aerobic (YL/G of 1.67 mol/mol) and the resulting constructs were cartilaginous. More aerobic conditions generally resulted in larger constructs containing higher amounts of cartilaginous tissue components, while anaerobic conditions suppressed chondrogenesis in 3D tissue constructs.     

Rosmarinic acid synthesis in transformed callus culture of Coleus blumei benth

Agrobacteria mediated Coleus blumei tumour tissues were cultured in vitro on MS medium. Sixteen diversified transformed callus cultures were maintained for several years in the absence of plant growth regulators and antibiotics without affecting the growth rate. Rosmarinic acid was detected spectrophotometrically in all tissue lines but in different quantities. The highest rosmarinic acid accumulation detected was 11% of dry tissue mass. The relation between culture growth and rosmarinic acid production was investigated in three callus lines. The lines showed different rosmarinic acid accumulation in relation to their growth rate; it was either parallel or inversely related to the tissue growth. The effects of certain medium constituents on the callus growth and rosmarinic acid accumulation were examined in four tumour cell lines. Addition of 4% or 5% sucrose stimulated rosmarinic acid synthesis and decreased callus growth. Nitrogen reduction to one half or one quarter of initial concentration did not affect rosmarinic acid synthesis and decreased callus growth in three lines, while it increased rosmarinic acid accumulation and callus growth in one line. Addition of 0.1 mg/l Phe stimulated rosmarinic acid production in two lines but had little effect on the rosmarinic acid level in others. Rosmarinic acid production was significantly improved on modified macronutrients, where the Ac2 line produced 16.5 mg of rosmarinic acid per tube (0.2 g of dry wt) after being in culture for 35 days.     

Protein flexibility and intrinsic disorder

Comparisons were made among four categories of protein flexibility: (1) low-B-factor ordered regions, (2) high-B-factor ordered regions, (3) short disordered regions, and (4) long disordered regions. Amino acid compositions of the four categories were found to be significantly different from each other, with high-B-factor ordered and short disordered regions being the most similar pair. The high-B-factor (flexible) ordered regions are characterized by a higher average flexibility index, higher average hydrophilicity, higher average absolute net charge, and higher total charge than disordered regions. The low-B-factor regions are significantly enriched in hydrophobic residues and depleted in the total number of charged residues compared to the other three categories. We examined the predictability of the high-B-factor regions and developed a predictor that discriminates between regions of low and high B-factors. This predictor achieved an accuracy of 70% and a correlation of 0.43 with experimental data, outperforming the 64% accuracy and 0.32 correlation of predictors based solely on flexibility indices. To further clarify the differences between short disordered regions and ordered regions, a predictor of short disordered regions was developed. Its relatively high accuracy of 81% indicates considerable differences between ordered and disordered regions. The distinctive amino acid biases of high-B-factor ordered regions, short disordered regions, and long disordered regions indicate that the sequence determinants for these flexibility categories differ from one another, whereas the significantly-greater-than-chance predictability of these categories from sequence suggest that flexible ordered regions, short disorder, and long disorder are, to a significant degree, encoded at the primary structure level.