The goal of neuromorphic engineering is to create electronic systems that model the behavior of biological neural systems. Neuromorphic systems can leverage a combination of analog and digital circuit design techniques to enable computational modeling, with orders of magnitude of reduction in size, weight, and power consumption compared to the traditional modeling approach based upon numerical integration. These benefits of neuromorphic modeling have the potential to facilitate neural modeling in resource-constrained research environments. Moreover, they will make it practical to use neural computation in the design of intelligent machines, including portable, battery-powered, and energy harvesting applications. Floating-gate transistor technology is a powerful tool for neuromorphic engineering because it allows dense implementation of synapses with nonvolatile storage of synaptic weights, cancellation of process mismatch, and reconfigurable system design. A novel neuromorphic hardware system, featuring compact and efficient channel-based model neurons and floating-gate transistor synapses, was developed. This system was used to model a variety of network topologies with up to 100 neurons. The networks were shown to possess computational capabilities such as spatio-temporal pattern generation and recognition, winner-take-all competition, bistable activity implementing a "volatile memory", and wavefront-based robotic path planning. Some canonical features of synaptic plasticity, such as potentiation of high frequency inputs and potentiation of correlated inputs in the presence of uncorrelated noise, were demonstrated. Preliminary results regarding formation of receptive fields were obtained. Several advances in enabling technologies, including methods for floating-gate transistor array programming, and the creation of a reconfigurable system for studying adaptation in floating-gate transistor circuits, were made.

We develop statistical methods for tackling two important problems in genetic association studies. First, we propose a Bayesian approach to overcome the winner's curse in genetic studies. Second, we consider a Bayesian latent variable model for analyzing longitudinal family data with pleiotropic phenotypes. Winner's curse in genetic association studies refers to the estimation bias of the reported odds ratios (OR) for an associated genetic variant from the initial discovery samples. It is a consequence of the sequential procedure in which the estimated effect of an associated genetic marker must first pass a stringent significance threshold. We propose a hierarchical Bayes method in which a spike-and-slab prior is used to account for the possibility that the significant test result may be due to chance. We examine the robustness of the method using different priors corresponding to different degrees of confidence in the testing results and propose a Bayesian model averaging procedure to combine estimates produced by different models. The Bayesian estimators yield smaller variance compared to the conditional likelihood estimator and outperform the latter in the low power studies. We investigate the performance of the method with simulations and applications to four real data examples. Pleiotropy occurs when a single genetic factor influences multiple quantitative or qualitative phenotypes, and it is present in many genetic studies of complex human traits. The longitudinal family studies combine the features of longitudinal studies in individuals and cross-sectional studies in families. Therefore, they provide more information about the genetic and environmental factors associated with the trait of interest. We propose a Bayesian latent variable modeling approach to model multiple phenotypes simultaneously in order to detect the pleiotropic effect and allow for longitudinal and/or family data. An efficient MCMC algorithm is developed to obtain the posterior samples by using hierarchical centering and parameter expansion techniques. We apply spike and slab prior methods to test whether the phenotypes are significantly associated with the latent disease status. We compute Bayes factors using path sampling and discuss their application in testing the significance of factor loadings and the indirect fixed effects. We examine the performance of our methods via extensive simulations and apply them to the blood pressure data from a genetic study of type 1 diabetes (T1D) complications.

Gamma oscillations of local field potentials (LFP) in the basal ganglia and thalamus had not been studied during saccades. Eleven patients were studied during deep brain stimulation (DBS); 6 were in the subthalamic nucleus (STN); 3 in the globus pallidus interna (GPi); and 2 in the thalamic ventralis intermedius nucleus (Vim). Patients performed horizontal saccades to visual targets while LFPs from DBS electrodes, scalp electroencephalogram (EEG), and electrooculogram (EOG) were recorded. Wavelet spectrograms were generated and saccade onset and event-related gamma synchronizations (ERS) were compared to baseline without eye motion. ERS were recorded at and after saccade onset in the STN, GPi and Vim, EEGs and EOGs; but were absent during target light illumination without saccades. ERS were symmetric in all DBS contacts and appeared identical in DBS LFPs, frontal EEGs and EOGs. These findings indicate their origin from extraocular muscle spike potentials rather than brain neural activity.

We develop statistical methods for tackling two important problems in genetic association studies. First, we propose a Bayesian approach to overcome the winner's curse in genetic studies. Second, we consider a Bayesian latent variable model for analyzing longitudinal family data with pleiotropic phenotypes. Winner's curse in genetic association studies refers to the estimation bias of the reported odds ratios (OR) for an associated genetic variant from the initial discovery samples. It is a consequence of the sequential procedure in which the estimated effect of an associated genetic marker must first pass a stringent significance threshold. We propose a hierarchical Bayes method in which a spike-and-slab prior is used to account for the possibility that the significant test result may be due to chance. We examine the robustness of the method using different priors corresponding to different degrees of confidence in the testing results and propose a Bayesian model averaging procedure to combine estimates produced by different models. The Bayesian estimators yield smaller variance compared to the conditional likelihood estimator and outperform the latter in the low power studies. We investigate the performance of the method with simulations and applications to four real data examples. Pleiotropy occurs when a single genetic factor influences multiple quantitative or qualitative phenotypes, and it is present in many genetic studies of complex human traits. The longitudinal family studies combine the features of longitudinal studies in individuals and cross-sectional studies in families. Therefore, they provide more information about the genetic and environmental factors associated with the trait of interest. We propose a Bayesian latent variable modeling approach to model multiple phenotypes simultaneously in order to detect the pleiotropic effect and allow for longitudinal and/or family data. An efficient MCMC algorithm is developed to obtain the posterior samples by using hierarchical centering and parameter expansion techniques. We apply spike and slab prior methods to test whether the phenotypes are significantly associated with the latent disease status. We compute Bayes factors using path sampling and discuss their application in testing the significance of factor loadings and the indirect fixed effects. We examine the performance of our methods via extensive simulations and apply them to the blood pressure data from a genetic study of type 1 diabetes (T1D) complications.

Gamma oscillations of local field potentials (LFP) in the basal ganglia and thalamus had not been studied during saccades. Eleven patients were studied during deep brain stimulation (DBS); 6 were in the subthalamic nucleus (STN); 3 in the globus pallidus interna (GPi); and 2 in the thalamic ventralis intermedius nucleus (Vim). Patients performed horizontal saccades to visual targets while LFPs from DBS electrodes, scalp electroencephalogram (EEG), and electrooculogram (EOG) were recorded. Wavelet spectrograms were generated and saccade onset and event-related gamma synchronizations (ERS) were compared to baseline without eye motion. ERS were recorded at and after saccade onset in the STN, GPi and Vim, EEGs and EOGs; but were absent during target light illumination without saccades. ERS were symmetric in all DBS contacts and appeared identical in DBS LFPs, frontal EEGs and EOGs. These findings indicate their origin from extraocular muscle spike potentials rather than brain neural activity.

Neurological diseases in newborns are usually first revealed by seizures, which are characterised by a synchronous discharge of a large number of neurons. Failure to control seizures may lead to brain damage or even death. The importance of this problem prompted many researchers to look for accurate automatic methods for seizure detection. Nonstationarity and multicomponent behaviour of newborn EEG signals made this task very challenging. The significant overlap in the characteristic of background and seizure activities in newborn EEG signals added to the difficulty of seizure detection. This research uses time-frequency based methods for automatic seizure detection. Since time-frequency signal analysis methods use joint representation in both time and frequency domains, they proved to be very suitable for analysis and processing of nonstationary and multicomponent signals such as newborn EEG. Before using any seizure detector, the EEG data is pre-processed in order to reduce the noise effects using a time-frequency based technique. The proposed method is based on the singular value decomposition (SVD) technique applied to the matrix representing the time-frequency distribution (TFD) of the EEG signal. It has been shown that by appropriately filtering the singular vectors associated with the TFD, one can effectively enhance the desired information embedded in the signal. Neonatal EEG seizures can have signatures in both low frequency (lower than 10 Hz) and high frequency (higher than 70 Hz) areas. The seizure detection techniques proposed in the literature concentrated on using either low frequency or high frequency signatures but not both simultaneously. These methods tend to miss the seizures that reveal themselves only in one of the two frequency areas. In this research, we propose a detection method that uses seizure features in both low and high frequency areas. To detect EEG seizures using the low frequency signatures, an SVD-based technique is employed. The technique uses the estimated distribution function of the singular vectors associated with the time-frequency distribution of EEG epochs to discriminate between seizure and nonseizure patterns. The high frequency signatures of seizures are mostly the result of spike events in the EEG signals. To detect these spike events, the signal is mapped into the TF domain. The high instantaneous energy of spikes is reflected as a localised energy in the high frequency area of the TF domain. Consequently, a spike can be seen as a ridge in this area of the TF domain. It has been shown that during seizure activity there is regularity in the distribution of the interspike intervals. This feature has been used as the basis for discriminating between seizure and nonseizure patterns. The performance results obtained by applying the proposed methods on EEG signals extracted from a number of newborns show the superiority of these methods over the existing ones.

In daily activities, hand use is dominated by individuated thumb and finger movements, and by grasping. This thesis focused on the level of ???independence??? of the digits and its relationship to hand grasps, from the level of the motor units to the level of synergistic grasping forces. Four major studies were conducted in healthy adult volunteers. First, spike-triggered averages of forces produced by single motor units in flexor pollicis longus (FPL) in a grasp posture showed small but significant loading of the index, but not other fingers. This reflected a neural rather than anatomical coupling, as intramuscular stimulation produced minimal effect in any finger. Also, FPL had a surprisingly large number of low-force motor units and this may account for the thumb???s exceptional dexterity and force stability compared with the fingers. Second, independent control of extensor digitorum (ED) was more limited than flexor digitorum profundus (FDP), as more ED motor units of a ???test??? finger were recruited inadvertently by extension than by flexion of adjacent digits. Third, ???force enslavement??? in maximal voluntary tasks was greater in digit extension than flexion. The distribution of force enslavement (and deficits) matched the pattern of daily use of the digits (alone and in combination), and reveals a neural control system which preferentially lifts fingers together from an object by extension but allows an individual digit to flex to contact an object so the finger pads can engage in exploration and grasping. Finally, during grasping, irrespective of whether a digit had been lifted from the object, coherence among forces generated by the digits was similar. In addition, the coherence between finger forces was independent of any contraction of the thumb, was stable over 2 months, and required no learning. The pattern of coherence between digital grasping forces may be closely related to the level of digit independence and daily use. Overall, the grasp synergy was remarkably invariant over the various tasks and over time. In summary, this thesis demonstrates novel aspects of the properties of FPL, the lack of complete independence of the digits, and robustness in the production of flexion forces in hand grasps.

Conselho Nacional de Desenvolvimento Científico e Tecnológico / The study aimed to evaluate the yield, chemical composition, yield and morphological characterization of seeds of annual ryegrass (Lolium multiflorum Lam) with two nitrogen sources: urea (nitrogen source most widely used and in Brazil, mainly in the south) and ammonium sulfate (source little used by producers, with virtually no loss by volatilization of nitrogen applied dressing), the nitrogen rate of zero, 100, 200, 300 and 400 kg ha- ¹, the agricultural year of 2009. Each treatment received one to three fresh cuts before the deferral for seed production in order to simulate a grazing situation, for ratings of productivity and chemical composition of pasture. The experimental design was randomized blocks. Based on the results did not differ between the sources used for the variables total dry matter, crude protein, acid detergent fiber, neutral detergent fiber, seed yield, number of spikelets per ear, ear length and spikelet number of seeds and length of seeds in three sections. The dose that gave the best response for most of the variables was 400 kg ha-¹ independent of nitrogen source used and the use of pasture with up to three cuts shows no decrease in quality and productivity of the same, but significantly affects the components of seed yield. Based on the variables may be concluded that there was no difference between the applied sources, then the ammonium sulfate had a similar response to urea can then be used as an alternative source of nitrogen. / O trabalho teve por finalidade avaliar a produtividade, composição bromatológica, rendimento e caracterização morfológica de sementes da forragem de azevém anual (Lolium multiflorum Lam.), com duas fontes de nitrogênio: uréia (fonte de nitrogênio mais difundida e utilizada no Brasil, principalmente no Sul do país) e sulfato de amônio (fonte pouco utilizada pelos produtores, com perda praticamente nula por volatilização de nitrogênio aplicado em cobertura), nas doses de nitrogênio (N) de zero, 100, 200, 300 e 400 Kg ha-¹, no ano agrícola de 2009. Cada tratamento recebeu de um a três cortes na matéria fresca antes do diferimento para a produção de sementes, a fim de simular uma situação de pastejo. O delineamento experimental adotado foi blocos ao acaso, seno quatro blocos e três repetições. Com base nos resultados obtidos não houve diferença estatística significativa entre as fontes de nitrogênio utilizadas para as variáveis: matéria seca total, proteína bruta, fibra detergente ácida, fibra detergente neutro, rendimento de sementes, número de espiguetas por espiga, comprimento de espiga e espigueta, número de sementes e comprimento de sementes. Em relação às doses, a dose de 400 Kg ha-¹ apresentou a melhor resposta para a maioria das variáveis estudadas e a utilização da pastagem com até três cortes não apresenta decréscimos na qualidade e produtividade da mesma, porém afeta significativamente os componentes do rendimento de sementes.

The Human Immunodeficiency Virus (HIV) is a member of the retroviridae family from lentivirus genus which primarily infects CD4+ T cells and also to lesser degree monocytes, macrophages, and dendritic cells causing progressive failure of the immune system, ultimately leading to development of acquired immunodeficiency syndrome (AIDS). Currently ~ 37 million people are infected with HIV-1 with approximately 2 million new infections occurring every year (UNAIDS, 2016). Developing safe, effective, and affordable vaccines to prevent HIV infection is the best hope for controlling the HIV/AIDS pandemic. Envelope glycoprotein (Env) on the HIV-1 virion surface is synthesized as a single precursor protein gp160 which is cleaved by furin to form the gp120 and gp41 subunits. gp41 is inserted into the membrane, while gp120 remains non-covalently associated with the ectodomain of gp41 to form a trimer of heterodimers. gp120 binds to the CD4 receptor on CD4+ T cells, which triggers a series of conformational changes leading to the exposure of co-receptor binding sites on gp120. Subsequent binding to the co-receptor (CXCR4 or CCR5) on T-cells initiates fusion of cellular and viral membranes via gp41 subunit. The envelope glycoprotein gp120, on the virion surface is the most accessible component of HIV-1 to the host immune system, and the target of most of the neutralization response. However, the virus has evolved many efficient ways to escape this immune surveillance. Extensive glycosylation of gp120 is one way by which it masks critical neutralization epitopes and the presence of immunodominant long variable loops focuses the immune response away from conserved regions. Certain conserved epitopes are cryptic and get exposed only after gp120 binds to its receptor. Also gp120 and gp41 are highly flexible molecules, attached in a non-covalent fashion to form a trimer of heterodimers, leading to inherent metastability of the Env. This results in exposure of a large number of non-native conformations to the immune system and thus minimizes elicitation of neutralizing antibodies. Despite these defense mechanisms, about 20-30% of HIV-1 patients do generate a broad neutralization response. Although these bNAbs and their epitopes have been identified, eliciting similar bNAbs through immunization is challenging. Monomeric gp120 when used as an immunogen elicits non neutralizing antibodies. This indicates that the epitopes of bNAbs are not present in the right conformation on this molecule. A rational design approach which focuses the immune response towards specific epitopes targeted by bNAbs is required, with the aim to maximize the exposure of conserved neutralization epitopes and to simultaneously ensure minimal exposure of variable non neutralizing epitopes. This can likely be achieved either by (a) stabilization of native Env trimers, or/and by (b) protein fragment design. Chapter 1 gives a brief description of HIV-1 virus. Structural features of the Env protein are described along with epitopes targeted by various bNAbs. Various strategies employed towards structure based vaccine design are discussed. One of the strategies towards rational vaccine design is using protein fragment based approaches. Grafting epitopes onto heterologous scaffolds is a promising approach which can provide more structural stability to the epitope, helps focus immune response on the epitope of interest and can be employed in a prime boost strategy for immunization studies. In a scaffold based approach we used crystal structure information of gp120 in complex with bNAb b12 to define the epitope of this antibody. In Chapter 2 we use this epitope information to graft the epitope on an unrelated scaffold protein to design unique epitope scaffolds. We report a computational strategy to graft the discontinuous epitope of b12 antibody onto different scaffold proteins. Our strategy focuses on identifying the best match of the target scaffold to the query protein so as to cause the least structural disturbance in the scaffold protein. The best hits were screened for binding to b12 using Yeast Surface Display (YSD). Random mutant libraries were also generated to screen for better b12 binders using YSD. We further characterized a few of these epitope scaffolds after purifying them from bacterial systems. One of the epitope scaffolds 1mkh_E2 bound to b12 with a KD value of 7.5µM. 2bodx_03, an unoptimized epitope scaffold reported previously (Azoitei et al, 2011) binds b12 with a KD value of 300μM. Thus our epitope scaffold 1mkh_E2 shows reasonable binding to b12 without any optimization. We are currently purifying other b12 epitope scaffolds and will be characterizing them for binding to b12. We have previously used a protein minimization strategy to design fragments of gp120, called b122a and b121a comprising a compact beta barrel on the lower part of the outer domain in order to focus the immune response towards the b12 epitope. (Bhattacharyya et al, 2013). These were bacterially expressed, found to be partially folded, however, could bind the broadly neutralizing antibody b12 with micromolar affinity. In rabbit immunization studies sera obtained following four primes with the b122a fragment protein and two boosts with full-length gp120 showed broad neutralization of a panel of multiple viruses across different clades (Bhattacharyya et al, 2013). In the present work, These designs were further stabilised by introducing various disulphides. One of the disulphide mutants b122a1-b showed better binding to b12 compared to b122a and increased protection to protease digestion. However these are partially structured as assessed by CD. In Chapter 3 we attempted to evolve stabilized versions of b122a1-b by using a genetic selection based on antibiotic resistance described previously (Foit et al, 2009). We were successfully able to show an in-vivo stability difference between b122a and b122a1-b. From the library generated in the background of b122a1-b using random mutagenesis, a few apparently stabilized mutants were isolated. Most of these mutations were hydrophobic to polar substitutions at exposed positions while a few of the mutations were substitutions with similar side chain chemistry as in wildtype. In future studies we will measure mutant stabilities and binding affinity to b12. A set of similar fragment immunogens were also designed based on subtype C CAP210 gp120 sequences. In Chapter 4 we describe various immunization studies comprising of different sets of b12 epitope based fragment immunogens. In one study we displayed some of these immunogens on Qβ VLPs. In another study, we tested subtype C based fragment immunogens. The humoral immune response was probed in terms of generation of antibodies against the immunogens using ELISA. Neutralization activity of the sera was measured in a standard TZM-bl assay. Sera raised against these particles in rabbit immunization studies could neutralize Tier1 viruses across different subtypes. The group primed with particles displaying b122a1-b and the group primed with b122a conjugated to particle in the presence of adjuvant contained significantly higher amounts of antibodies directed towards the CD4bs than sera from the group primed with empty particles and boosted with gp120. This study demonstrates the overall utility of the particle based display approach. In immunization studies with subtype C derived fragment immunogens as primes, no significant neutralization was seen even for Tier 1 viruses. In this study, the group primed and boosted with full length gp120 performed better than other groups suggesting that antibodies elicited against regions present in these subtype C priming immunogens are non-neutralizing. One of the rational vaccine design strategies is by stabilization of native Env trimers. In previous studies, a disulfide bond was engineered between gp120 and gp41 of Env to stabilize the interactions (SOS gp140). An I559P mutation was also introduced to stabilize the native gp41 conformation in the context of disulfide engineered Env (SOSIP gp140). The purified, soluble SOSIP gp140 immunogens were trimeric and cleaved properly and are believed to be one of the closest mimics of native Env trimers. However, these immunogens have so far failed to elicit broad neutralizing responses. In Chapter 5, we use structural information derived from high resolution atomic structure of native like cleaved gp140 BG505-SOSIP, to provide an alternate strategy to form uncleaved trimeric gp140s by cyclic permutation to design molecules that mimic cleaved trimers. The structure reveals that the gp41 C-terminus is in very close proximity (~8Å) to the N-terminus of gp120 from an adjacent subunit. We have designed a cyclic permutant of gp140 from JRFL strain where the gp41 C terminus is now connected to the gp120 N-terminus with a short linker. This novel connectivity results in preservation of the native gp41 N-terminus along with a much shorter linker length than in conventional gp140. This might promote trimer folding and stabilization because of the resulting decreased magnitude of conformational entropy change during folding. The structure also reveals that the gp120 C-terminus is close to the trimer axis, and due to cyclic permutation, this becomes the new C-terminus of gp140. To further stabilize the trimeric form, we have attached a foldon trimerization domain at the C terminus. The protein has been expressed and purified from mammalian cells. The protein exists primarily as a trimer in solution as assessed by SEC-MALS. It shows better binding to broadly neutralizing antibody b12 when compared to b6, a non-neutralizing antibody. Further biophysical characterization of the protein is in progress. We have previously described design of a bacterially expressed outer domain derivative of gp120 (ODEC) that had V1/V2 and V3 loops deleted and bound CD4 (Bhattacharyya et al, 2010). To improve the initial ODEC design, three different rational design strategies were used. In the first approach, residue frequency based methods were used to design a construct named ODECConsensus. In another approach, a cyclic permutant of ODEC (CycV4OD) was designed with new N and C termini in the flexible V4 loop. In the third approach the bridging sheet (BS) region was deleted from ODEC to form ODECΔBS. In Chapter 6 we have used hydrogen deuterium exchange-mass spectrometric analysis (HDX-MS) to study conformational flexibility of these fragment immunogens. These studies revealed that all the three immunogens show reduced conformational flexibility compared to ODEC. 5-7 protons remain protected up to 2 hours whereas for ODEC, exchange completes at 20 minutes. This reduced flexibility correlates with 6-20 fold tighter VRC01 binding relative to ODEC. In rabbit immunizations, all three constructs elicit significant gp120 titers as early as week 6 in the absence of any gp120 boost whereas ODEC shows significant gp120 titers only after two gp120 boosts. Week 24 sera elicited after immunization with ODECΔBS, ODECConsensus and CycV4OD boosted with gp120 show neutralization of multiple Tier 1 viruses from subtype B and C, whereas corresponding ODEC immunized animals failed to show a neutralizing response. This study demonstrates that reduced conformational flexibility correlates with better antigenicity and an improved immunogenicity profile for these fragment immunogens. Also we have used HDX-MS studies to one of the stem based HA fragment immunogen pH1HA10-foldon described previously (Mallajosyula et al, 2014) to do peptide finger printing and find regions of protein showing increased protection to hydrogen deuterium exchange and thus derive some structural insights about this trimeric fragment immunogen. Peptide mapping experiments show that the HA stem fragment peptides are exchanging rapidly with more than 90% exchange completing by 30 s for most of the peptides. The well folded foldon trimerization domain peptide shows a very slow exchange profile. A few of the HA peptides exchange slowly with 1-2 protons exchanging after 30 s. Fast exchange seen for this fragment immunogen may be due to truncation of the stem region leading to greater solvent accessibility of the trimer interface.

Conselho Nacional de Desenvolvimento Científico e Tecnológico / Crop simulation models are important tools to help farmers in planning management practises and flowering time of cut flowers, like gladiolus (Gladiolus x grandiflorus Hort.). The objective of this study was to develop a robust gladiolus phenology model, named PhenoGlad, for field applications. The model describes the timing of developmental stages, including harvest point, the vase life of gladiolus spikes and the low (chilling) and high (heat) temperature effects on spike quality. The gladiolus developmental model simulates gladiolus phenology using a non-linear temperature response function and, by accumulating daily development rates considering three main phases: corms sprouting phase, vegetative phase, and reproductive phase. Data from nine field experiments conducted during five years (2011 2015) in three locations across the Rio Grande do Sul State and in one location in Santa Catarina State, Brazil, were used. These cultivar x planting dates x years x locations experiments provide a rich data set for calibrating and evaluating the gladiolus model. The PhenoGlad model accurately simulated the dynamics of leaf development, final leaf number and the timing of developmental stages using genotype-specific coefficients that can be estimated from thermal time. The performance of the model was improved when the simulations started from emergence compared to when simulations started at the planting date. PhenoGlad showed good stability among cultivars, planting dates, years and sites, with an RMSE of 0.5 leaves for leaf development and final leaf number, 6.5 to 5.0 days for the date of reproductive developmental stages when the model started from planting or from emergence, respectively, and 1.3 days for simulating the vase life of harvested spikes. PhenoGlad was also efficient in predicting the effects of chilling and high temperatures damage on florets. / Modelos matemáticos de simulação das culturas agrícolas são importantes ferramentas para auxiliar os produtores a planejar práticas de manejos a campo e prever a data de florescimento de flores de corte, como o gladíolo (Gladiolus x grandiflorus Hort.). O objetivo deste trabalho foi desenvolver um modelo robusto de fenologia chamado PhenoGlad com aplicações de campo. O modelo descreve a data de ocorrência dos estágios de desenvolvimento da cultura do gladíolo, incluindo o ponto de colheita, a vida de vaso de hastes de gladíolo e os efeitos de baixas e altas temperaturas na qualidade das espigas. O modelo de desenvolvimento de gladíolo simula a fenologia da cultura utilizando uma função de resposta não-linear à temperatura, através do acúmulo de valores diários da taxa de desenvolvimento a partir do plantio ou da emergência, usando três principais fases de desenvolvimento da cultura: fase de brotação dos cormos, fase vegetativa e fase reprodutiva. Dados de nove experimentos de campo conduzidos durante cinco anos (2011 2015) em três locais do estado do Rio Grande do Sul e em um local em Santa Catarina, Brasil, foram utilizados. Estes experimentos com diferentes cultivares x datas de plantio x anos x locais proporcionam um rico banco de dados para calibração e validação do modelo para a cultura do gladíolo. O modelo PhenoGlad simulou satisfatoriamente a dinâmica de emissão de folhas, número final de folhas e a data dos estágios de desenvolvimento em gladíolo usando coeficientes genético-específicos que podem também ser estimados através da soma térmica. O modelo PhenoGlad mostrou uma boa estabilidade entre cultivares, datas de plantio, anos e locais, demostrado pelos baixos valores de RMSE (0,5 dias para aparecimento de folhas e número final de folhas, 6,5 a 5,0 dias quando o modelo é rodado a partir do plantio e da emergência, respectivamente, e 1,3 dias para simular a vida de vaso de espigas de gladíolo). O modelo PhenoGlad foi também eficiente em predizer os efeitos de baixas e altas temperaturas na qualidade das espigas.