Piloting an approach to rapid and automated assessment of a new research initiative: Application to the National Cancer Institute's Provocative Questions initiative.

Funders of biomedical research are often challenged to understand how a new funding initiative fits within the agency's portfolio and the larger research community. While traditional assessment relies on retrospective review by subject matter experts, it is now feasible to design portfolio assessment and gap analysis tools leveraging administrative and grant application data that can be used for early and continued analysis. We piloted such methods on the National Cancer Institute's Provocative Questions (PQ) initiative to address key questions regarding diversity of applicants; whether applicants were proposing new avenues of research; and whether grant applications were filling portfolio gaps. For the latter two questions, we defined measurements called focus shift and relevance, respectively, based on text similarity scoring. We demonstrate that two types of applicants were attracted by the PQs at rates greater than or on par with the general National Cancer Institute applicant pool: those with clinical degrees and new investigators. Focus shift scores tended to be relatively low, with applicants not straying far from previous research, but the majority of applications were found to be relevant to the PQ the application was addressing. Sensitivity to comparison text and inability to distinguish subtle scientific nuances are the primary limitations of our automated approaches based on text similarity, potentially biasing relevance and focus shift measurements. We also discuss potential uses of the relevance and focus shift measures including the design of outcome evaluations, though further experimentation and refinement are needed for a fuller understanding of these measures before broad application.

AM1 parameters for the prediction of 1H and 13C NMR chemical shifts in proteins.

The semiempirical quantum mechanical description of NMR chemical shifts has been implemented at the AM1 level with NMR-specific parameters to reproduce experimental (1)H and (13)C NMR chemical shifts. The methodology adopted here is formally the same as that of the previously published finite perturbation theory GIAO-MNDO-NMR approach [Wang, B.; et al. J. Chem. Phys. 2004, 120, 24.]. The primary impetus for this parametrization was the accurate capture of chemical environments of atoms in biological systems. Protein-specific parameters were developed on a training set that comprised five globular protein systems with varied secondary structure and a range in size from 46-61 amino acid residues. A separate set of parameters was developed using a training set of small organic compounds with an emphasis on functional groups that are relevant to biological studies. Our approach can be employed using semiempirical (AM1) geometries and can be executed at a fraction of the cost of ab initio and DFT methods, thus providing an attractive option for the computational NMR studies of much larger protein systems. Analysis carried out on 3340 (1)H and 2233 (13)C chemical shifts for protein systems shows significant improvement over the standard AM1 parameters. Using (1)H and (13)C specific parameters, the rms errors are from 1.05 and 21.28 ppm to 0.62 and 4.83 ppm for hydrogen and carbon, respectively.

MNDO parameters for the prediction of 19F NMR chemical shifts in biologically relevant compounds.

The semiempirical MNDO methodology for qualitative description NMR chemical shifts has now been extended with the addition of NMR-specific parameters for the fluorine atom. This approach can be employed using semiempirical (AM1/PM3) geometries with good accuracy and can be executed at a fraction of the cost of ab initio and DFT methods, providing an attractive option for the computational studies of (19)F NMR for much larger systems. The data set used in the parametrization is large and diverse and specifically geared toward biologically relevant compounds. The new parameters are applicable to fluorine atoms involved in carbon-fluorine bonds. These parameters yield results comparable to NMR calculations performed at the DFT (B3LYP) level using the 6-31++G(d,p) basis set. The average R (2) and rms error for this data set is 0.94 and 13.85 ppm, respectively, compared to 0.96 and 10.45 ppm when DFT methods are used.