email: as86 at

My C.V. - C.V.


I was born and raised in Santa Bárbara d’Oeste, just over an hour outside São Paulo city in Brazil. I attended the University of Michigan for my undergraduate degree from 2006 to 2010, where I graduated with a B.S. in Mathematics with emphasis in mathematical biology. Following graduation I swam professionally in 2011 and 2012. Afterwards, in order to further pursue applied mathematics and computational modeling, I enrolled at Rice University in the fall of 2012 where I am currently a PhD candidate in Dr. Amina Qutub’s lab.



My current work focuses on integrating omics data into constraint based models. I am developing novel  methods for the identification of (1) metabolic adaptations of deer mice to hypoxic conditions and (2) metabolic differences between cancer cells and healthy tissues. I am currently working on developing new methods for genome-wide metabolic reconstructions validation and analysis in humans, as well as clustering and analyzing expression data.


  • A. Schultz, S. Mehta, C.W. Hu, F.W. Hoff, T.M. Horton, S.M. Kornblau, A.A. Qutub (2016) Identifying Cancer Specific Metabolic Signatures Using Constraint Based Models. Pac Symp Biocomput. (link) (Supplemental Information)

  • A. Schultz, A.A. Qutub (2016) Reconstruction of Tissue-Specific Metabolic Networks Using CORDA, PLoS Comput Biol, 12(3). (link)

  • D.P. Noren, B.L. Long, R. Norel, K. Rrhissorrakrai, K. Hess, C.W. Hu, A.J. Bisberg, A. Schultz, E. Engquist, L. Liu, X. Lin, G.M. Chen, H. Xie, G.A.M. Hunter, P.C. Boutros, O. Stepanov, DREAM 9 AML-OPC Consortium, T. Norman, S.H. Friend, G. Stolovitzky, S. Kornblau, A.A. Qutub (2016) A Crowdsourcing Approach to Developing and Assessing Prediction Algorithms for AML Prognosis, PLoS Comput Biol, 12 (6). (link)

  • A. Schultz, A.A. Qutub (2015). Predicting internal cell fluxes at sub-optimal growth. BMC Syst Biol, 9(1), 18. (link)


The COnstraint Based Reconstruction and Analysis toolbox (Schellenberger,, 2012, Nat Protocol, 6(9) 1290-1307), or COBRA, is a widely used toolbox for the reconstruction, visualization and analysis of Genome-Wide Metabolic Resonstructions. Here I am sharing a few functions I have developed that can be implemented with this toolbox. Reconstructions for several organisms can be directly downloaded from the Biochemical, Genetic and Genomic database (Schellenberger,,2010, BMC Bioinformatics, 11:213), or BiGG ( The latest, most widely used human metabolic reconstructions are Recon2 (Thiele et. al., 2013, Nature biotechnology, 31(5)), which can be found at, and HMR2 (Mardinoglu, et al., 2014, Nature communications, 5), which can be found at Please feel free to email me with any bugs, questions, or general comments.

Monte-Carlo Sampling


mfACHR – A faster MATLAB implementation of the ACHR algorithm introduced and validated in Schultz et. al., 2016, Pac Symp Biocomput (manuscript accepted). This function moves all sampled point at the same time, taking advantage of the efficiency of MATLAB’s matrix operations.

FCAcompress – This function uses the Flux Coupling Analysis F2C2 toolbox (Larhlimi 2012) to combine fully coupled reactions into a single reaction, reducing the size of the metabolic model. This model reduction then makes the sampling procedure 40-45% faster in the two small models tested here. While the model is not reduced is dimension, and the same number of steps is needed for convergence, each step is made faster computationally since the stoichiometric matrix is smaller. A function that returns the flux distributions sampled in the compressed model to the original model is also provided. A tutorial showing how to use these functions and validating this method is also provided.

Tissue-Specific Models

CORDA – This function generates context-specific metabolic models of metabolism based on a generalized reconstruction and experimental data. CORDA includes a flexible and a non-flexible reactions core, and takes a non-parsimonious approach to the reconstruction process. When compared to previous, similar methods, CORDA agrees better with experimental data, performs fast computationally, and demonstrates better model functionality and capacity. This method was introduced and validated in Schultz and Qutub, 2016, PLoS Comput Biol, 12(6).

CORDA2 – A faster and deterministic version of CORDA introduced in Schultz et. al., 2016, Pac Symp Biocomput (manuscript accepted). This version of CORDA takes a noise-independent approach to the tissue-building algorithm, giving the same output every function iteration. CORDA2 and CORDA models were shown to be very similar.

Flux-Balance Analysis

corsoFBA – Function predicts a cost reduced flux distribution at a pre-defined percentage from optima. Here, a cost is first associated with each reaction in the network. Next, the objective function is fixed at a predetermined value, and this cost is optimized throughout the model, thus predicting a cost-optimal flux distribution. This function was used in Schultz and Qutub, 2015, BMC Syst Biol, 9(1), 18 to predict internal cell fluxes in E. coli.

corsoFBA2 – Added functionality to corsoFBA. Allows user to constraint values by both percentage and absolute value, defines more default values, and model decomposition is performed faster.

Pathway Analysis

expa_elementary – This function allows the user to calculate Fundamental Pathways as described in Schultz and Qutub, 2015, BMC Syst Biol, 9(1), 18. These are similar to Extreme Pathways, but are calculated in a model with no currency metabolites. With that, fewer pathways are obtained and a currency metabolite imbalance is associated with each pathway, which is also returned by this function. This function first helps guide the user as to which currency metabolites to delete, then uses expa to calculate the extreme pathways.

expa – This function implements the algorithm described by Schilling et. al. (Schilling et. al., 2000, Journal of theoretical biology, 203(3)) to calculate the Extreme Pathways of a metabolic model. This algorithm is relatively expensive and the number of Extreme Pathways increases drastically with network size. Therefore, this function is meant to be used only with relatively small networks.


Metabolic Maps

jsontomap – Function converts json files downloaded from Escher (King, et. al. 2015) to map structs that can be used with the COBRA toolbox. A tutorial on how to use this function is included with the download.

getrxnstp -This function allows you to modify maps in the COBRA toolbox to include only reactions you want. A tutorial on how to use the function is included.

drawFluxNoText – These functions provide a simple modification to the COBRA functions that plot metabolic maps, allowing them to plot without the reaction and metabolite names. This makes the map less crowded and faster to plot. A tutorial is included as well.

Keeping Up With the Literature

Keeping up with the literature within your field can be difficult and time consuming. I have found that NCBI alerts are good, but are limited to searches by authors and search terms, and do not consider a library of papers you may have. These alerts can also send several emails daily if you set too many. I have written some R code that searches new papers daily based on authors and search terms, and emails those results daily in a single email. This code also searches papers that cite several papers in a library you have defined, and suggests highly related papers based on citations. The files to do this, along with a tutorial on how to set it up, can be downloaded here. The tutorial is meant to make the setup easy even if you have never programmed before. Please send along any bugs, difficulties using or setting up, and any suggestions you may have on how to make this better. I am trying to make this code as bulletproof and friendly as possible.


I attended the University of Michigan on an athletic scholarship, where I was part of the swim team. Visit my MGoBlue Profile for a complete list of accomplishments. I was also a member of the Brazilian national team and swam professionally back in Brazil.