No abstract

We present QSPR models for normal boiling points employing a neural network approach and descriptors calculated using semiempirical MO theory (AM1 and PM3). These models are based on a data set of 6000 compounds with widely varying functionality and should therefore be applicable to a diverse range of systems. We include cross-validation by simultaneously training 10 different networks, each with different training and test sets. The predicted boiling point is given by the mean of the 10 results, and the individual error of each compound is related to the standard deviation of these predictions. For our best model we find that the standard deviation of the training error is 16.5 K for 6000 compounds and the correlation coefficient (R2) between our prediction and experiment is 0.96. We also examine the effect of different conformations and tautomerism on our calculated results. Large deviations between our predictions and experiment can generally be explained by experimental errors or problems with the semiempirical methods.

It is known that proteins can adopt different folds while sharing similar features for recognition of similar substrates or ligands, for example, in the binding sites of enzyme cofactors such as ATP. On the other hand, proteins that have highly flexible binding sites or belong to large and diverse protein families can bind structurally dissimilar ligands, as, for example, in the case of the matrix metalloprotease family. We have developed a database, PDBLIG, that classifies protein domains and ligands. The information stored includes each protein's function, domain class(es), which ligand(s) it binds, and so on. The database can provide valuable knowledge for drug discovery, supporting the answering of questions such as whether the same drug molecule can bind different target protein families and whether these families are related functionally or structurally, which ligand classes (such as metabolites or organic molecules) bind to a particular protein family and whether the ligands are druglike, and which target families bind a wide variety of ligands and whether different ligands are associated with different subfamilies.

Study of RgS− and RgS (Rg = Ne, Ar, and Kr) via slow photoelectron velocity-map imaging spectroscopy and ab initio calculationsAb initio potential energy surfaces have been constructed and used to carry out classical simulations of the reactions of X with HOC ϩ and of XH ϩ with CO ͑XϭNe, Ar, and Kr͒. The competition between rearrangement, XϩHOC ϩ →OCH ϩ ϩX, and abstraction, XϩHOC ϩ →XH ϩ ϩCO, has been examined, and found to favor abstraction in the cases where both processes are energetically allowed. The reaction of XH ϩ with CO is found to produce highly vibrationally excited ͓CHO͔ ϩ products.

High-level ab initio calculations using the G2** and G2(ZPE=MP2) methods have been employed to examine the effect of interaction with a range of neutral molecules (X) on the barrier to the degenerate proton-transport reactions in NNH+ (X = Ar, HF, CO, N2, H2O) and methyl cation-transport reactions in NNCH3 + (X = HF, H2, N2, HCl, H2O). It is found that the barriers to both proton and methyl cation transport are lowered from their values in the isolated ions of 182 and 152 kJ mol-1, respectively, by interaction with species having values of the proton or methyl cation affinity, respectively, lower than that of molecular nitrogen. Interaction with species that have larger values of proton or methyl cation affinities leads to a further lowering of the barrier, but transfer to the neutral molecule X becomes the more energetically favorable process in such cases. It is found that the ideal catalyst for ion transport should have an ion affinity close to but less than that of molecular nitrogen and have a large dipole moment.