RESUMO
Interactions between catalysts and substrates can be highly complex and dynamic, often complicating the development of models to either predict or understand such processes. A dirhodium(II)-catalyzed C-H insertion of donor/donor carbenes into 2-alkoxybenzophenone substrates to form benzodihydrofurans was selected as a model system to explore nonlinear methods to achieve a mechanistic understanding. We found that the application of traditional methods of multivariate linear regression (MLR) correlating DFT-derived descriptors of catalysts and substrates leads to poorly performing models. This inspired the introduction of nonlinear descriptor relationships into modeling by applying the sure independence screening and sparsifying operator (SISSO) algorithm. Based on SISSO-generated descriptors, a high-performing MLR model was identified that predicts external validation points well. Mechanistic interpretation was aided by the deconstruction of feature relationships using chemical space maps, decision trees, and linear descriptors. Substrates were found to have a strong dependence on steric effects for determining their innate cyclization selectivity preferences. Catalyst reactive site features can then be matched to product features to tune or override the resultant diastereoselectivity within the substrate-dictated ranges. This case study presents a method for understanding complex interactions often encountered in catalysis by using nonlinear modeling methods and linear deconvolution by pattern recognition.
RESUMO
INTRODUCTION Cystic fibrosis is a multisystem autosomal recessive disease with wide variability in clinical severity. It is incurable and characterized by elevated and premature mortality, as well as poor quality of life. Its frequency, lethality and devastating impact on both the physical and psychological wellbeing of patients and their families, make it a serious health problem. Its frequency in Cuba is 1 in 9862 live births, where marked molecular heterogeneity of the CFTR gene makes molecular diagnosis difficult. Six mutations have been identified that together enable molecular characterization of only 55.5% of cystic fibrosis chromosomes. This paper presents national results of antenatal diagnostic testing, using direct and indirect methods, for detection of cystic fibrosis. OBJECTIVE Characterize the Cuban public health system's experience with antenatal molecular testing for cystic fibrosis from 1988 through 2011. METHODS A retrospective descriptive study was conducted with results of antenatal diagnostic testing of amniotic fluid, performed nationwide from 1988 through 2011, for 108 fetuses of couples with some risk of having children affected by cystic fibrosis, who requested testing. Polymerase chain reaction detected mutations p.F508del, p.G542X, p.R1162X, p.R334W, p.R553X and c.3120+1G>A, and markers XV2C and KM19. Data were analyzed using absolute frequencies and percentages, and presented in tables. RESULTS For 93 cases (86.1%), testing for cystic fibrosis was done using direct analysis of mutations p.F508del, p.G542X, p.R1162X, p.R334W, p.R553X and c.3120+1G>A; five cases (4.6%) were tested indirectly using markers XV2C/Taq I and KM19/Pst I; and 10 (9.3%) were tested using a combination of the two methods. A total of 72 diagnoses (66.7% of studies done) were concluded, of which there were 20 healthy fetuses, 16 affected, 27 carrier, and 9 who were either healthy or carriers of an unknown mutation. CONCLUSIONS Direct or indirect molecular study was successfully used in over half of antenatal tests requested by couples throughout Cuba at risk of having children affected by cystic fibrosis, which is of great social value because of CF's burden on affected persons and their families.