Cerebral palsy (CP) is a progressive neuromuscular disorder caused by a non-progressive damage to the brain during child birth or neonatal stage of life. Approximately two to four in 1000 newborns per year and over 17 million children worldwide are affected by CP, which results in motor function deficiency and subsequent gait impairment. Muscle conditions in individuals with CP is highly heterogeneous and characterised by alterations of musculoskeletal structure, motor control, as well as hypertonia, spasticity, contracture and altered muscle composition, making planning of appropriate treatments an unresolved challenge. Clinical examination, clinical history, and three-dimensional gait analysis are generally used to inform treatment planning; however, treatment outcomes to correct deformities, stabilise joints, and balance muscle power, remain poor to moderate. However, computational tools able to assess the internal biomechanics of an individual with CP might aid the treatment planning, including surgery. Neuromusculoskeletal (NMSK) models are a digital representation of the structure and physiology of the human neural, muscular, and skeletal systems. Using data obtained from gait analysis and medical imaging, NMSK models can estimate the internal loads (i.e., muscle and joint contact forces) acting on selected joints and tissues. These data may be used, first, to inform clinical decision making, and second, to predict alteration in gait post-surgery. Specifically, altered joint loading could lead to bone deformities and subsequent decline in gait quality even in asymptomatic patients, and as such early detection of altered internal biomechanics could provide a potent screening tool. Nevertheless, including NMSK model predictions to the clinical decision-making process has some challenges. Accurate estimation of muscle and joint contact forces requires collecting electromyograms (EMG) from numerous muscle sites during gait analysis. However, recording EMG data from a large set of muscles in children with CP is difficult due to excessive atrophy and small size of the lower limbs. Also, uncomfortably long data collection procedures may distress children with CP, which can prematurely end a data collection session. Fast, but effective, EMG data collection is always preferred in clinical settings, and consequently, EMGs are commonly recorded from four to five muscles in clinical gait analyses of the children with CP. This makes reliable and effective extrapolative estimation of unmeasured muscle excitations crucial to undertake NMSK modelling in clinical settings. Estimation of unmeasured muscle excitations may be facilitated using “muscle synergies”, which are the coordinated excitation of groups of muscles that are used to generate any rhythmic task (e.g., walking). Muscle synergies extracted from a group of processed EMG signals (known as measured excitations) may provide a tool to quantify motor control by finding low dimensional weighted “excitation” modules that are blended to generate the activity. Generally, there are two components of muscle synergies: synergy weights and excitation primitives, of which a linear combination produces a reconstructed set of muscle excitations. Individuals with CP exhibit muscle excitations and synergies with reduced complexity compared to their healthy counterparts, which means that synergies may be reliably extracted from the four to five EMG recordings generally collected during gait analysis of the children with CP. Given that NMSK modelling requires as input excitation data from a large set of muscles involved in movement generation, the application of muscle synergies has two-fold advantages: (i) muscle synergies could be used to estimate unmeasured muscle excitations using a small set of experimental EMG data collected in gait clinics, that (ii) in turn may potentially inform NMSK modelling to estimate internal biomechanics of muscles and joints during locomotion in children with CP. Previous studies have employed muscle synergies to inform NMSK models. However, muscle synergies have typically been identified from a small set of experimental EMG data that might not represent dynamic function of the NMSK system where a large number of muscles are involved. Usually a large set of muscle excitations (30 to 50) enabling a NMSK model, that are estimated and/or adjusted to well track experimental joint moments and EMGs when employing EMG-informed NMSK modelling, are referred to as dynamically consistent. Consequently, muscle synergies extracted from the set of dynamically consistent muscle excitations may be more neurophysiologically representative than those extracted from a small set of experimentally acquired EMG data. In this thesis the overarching goal was to develop a muscle synergy-informed NMSK modelling workflow where muscle synergies were extracted from a set of dynamically consistent muscle excitations with most appropriate synergy extraction method. Using such an approach, it was aimed to use these synergy-informed models to predict internal joint biomechanics of children with CP using minimal set of EMG recordings. Generally, the factorisation methods used to extract muscle synergies are based on strict assumptions about the data’s underlying probability distribution or within-data correlations. These assumptions, may be restrictive, and even questionable, for synergy analysis of EMGs, which a priori sets the data’s information content. That is, EMG data reflect information on the organisation of muscle excitations to generate movement, and probability distributions of EMG changes (or any data for that matter) reflecting this information content. Subsequently, factorisation methods with a priori assumption on strict distributions may generate imperfect muscle synergies unless the variation of the distribution is properly tracked. So, the neural information within varying EMG signals could also be assessed by the concomitant changes in the probability density function. As such, metrics based on the probability density function of EMG data may echo the information conveyed by the EMG signals before and after processing, and by factorisation, therefore assessing how much information was retained. Novel metrics based on the probability density function could be used to evaluate the best factorisation method that can track the change in distribution of EMG data, which would produce the best muscle synergies to represent the neural information content. Subsequently, the first study assessed four factorisation methods commonly used to extract muscle synergies from muscle excitations. Although there are many factorisation methods available for extracting muscle synergies, it is not clear why one method should be chosen over another. The selection of the factorisation methods is likely based on their availability in popular computer programs rather than theory underpinning the synergy extraction process. Assessing factorisation methods based on the EMG data’s probability distribution, a common underpinning assumption of any method, has not yet been received much attention. In this study, the four most common factorisation methods, i.e., non-negative matrix factorisation method, principal component analysis, independent component analysis, and factor analysis, were evaluated using “variance accounted for” and the similarities in the probability density function of the muscle excitations and corresponding muscle synergies. Consistent with previous work, it was found that the probability density functions of EMG data varied with locomotion speed. Furthermore, the probability distribution of EMG data is a priori assumed in most factorisation approaches, which is not compliant with this variation in probability density functions. However, the non-negative matrix factorisation was found to track the probability density function variations better than the other methods, since mathematically it did not depend on any underlying form of the EMG data’s probability distribution. As such, non-negative iv matrix factorisation method was chosen to extract muscle synergies from the muscle excitations in two further studies in this thesis. This study was published as Rabbi, M. F., Pizzolato, C., Lloyd, D. G., Carty, C. P., Devaprakash, D., & Diamond, L. E. (2020). Non-negative matrix factorisation is the most appropriate method for extraction of muscle synergies in walking and running. Scientific reports, 10: 8266. The second study investigated the feasibility of estimating unmeasured muscle excitations of children with CP using EMG collected from typically developing (TD) children. It has been previously shown, and was replicated in this study, that muscle synergies identified in children with CP are a simplified version, and a subset, of those synergies (particularly excitation primitives) found in their TD counterparts. Using this fundamental finding, this study developed and evaluated a muscle synergy extrapolation method to estimate unmeasured muscle excitations in children with CP using a large set of EMG data collected from TD participants. The larger set of TD EMG linear envelopes (measured excitations) were mapped onto the CP excitation primitives using pseudoinverse least squares to produce a larger hybrid set of synergy weightings. We identified the excitation primitives extracted from the best small set of selected muscle measured excitations from children with CP. It was found that measured excitations from three to four muscles in these children were able to estimate six to seven unmeasured muscle excitations with an acceptable level of variance accounted for. Furthermore, to ensure the information content was preserved in the reconstructed muscle excitations, the variation in probability density function was also compared between the experimental and estimated muscle excitations. It was found that the probability density function in reconstructed muscle excitations were similar to that in the original muscle excitations when using muscle synergy extrapolation method. Therefore, muscle synergy extrapolation method could facilitate the estimation of unmeasured muscle excitations of children with CP in clinical settings. All findings and results of this study have been published as Rabbi, M. F., Diamond, L. E., Carty, C. P., Lloyd, D. G., Davico, G., & Pizzolato, C. (2022). A muscle synergy-based method to estimate muscle activation patterns of children with cerebral palsy using data collected from typically developing children. Scientific reports, 12: 3599. The third study investigated if a synergy extrapolation method could be developed and incorporated with NMSK modelling to estimate the internal musculoskeletal biomechanics of children with CP with a minimal number of EMG recordings. Since a limited number of EMGs are collected in gait clinics, a comprehensive gait evaluation v using NMSK modelling might not be possible, the later which needs a large set muscle excitations as inputs. A modelling workflow is thus required to estimate the measured and unmeasured muscle excitations and internal biomechanics at the same time using only a small number of EMG recordings. Indeed, EMG-informed NMSK modelling that is limited by the number of unmeasured EMG recordings might not well estimate joint moments and contact forces. So, muscle synergies could inform such NMSK modelling to estimate unmeasured muscle excitations and subsequent internal joint biomechanics. However, the muscle synergies and excitations must be dynamically consistent, i.e., able to drive the NMSK model to well track experimental joint moments and measured EMG. To this end, since it is easier to collect EMG from a large set of muscles from TD children, synergies extracted from a healthy EMG set and an EMG-informed modelling was used to develop a synergy-informed NMSK modelling. From a minimal number of experimental EMG recordings, the synergy-informed NMSK models were able to estimate the muscle excitations and forces, and joint moments, as accurately as the current best EMG-assisted modelling, and outperformed static optimisation method. Research findings of this study will be submitted for publication as Rabbi, M. F., Davico, G., Diamond, L. E., Lloyd, D. G., Carty, C. P., & Pizzolato, C. (2022). Muscle synergy-informed neuromusculoskeletal modelling for children with cerebral palsy. Frontiers in Physiology, to be submitted. In conclusion, a muscle synergy-informed NMSK modelling workflow was developed where internal biomechanics for children with CP was estimated with a minimal set of experimental EMG data. While developing the workflow, non-negative matrix factorisation was found to be the most appropriate factorisation method to extract muscle synergies during locomotion, based similarities to the EMGs probability density function. Further, it was demonstrated how muscle synergies could be used to estimate unmeasured muscle excitations. Finally, a muscle synergy-informed NMSK model using EMG from only three or four muscles was developed to estimate muscle forces, joint moments and knee contact forces for children with CP during walking. Considering advantages in clinical data collection and in vivo estimation of internal biomechanics it appears that combining synergy-informed NMSK modelling and EMG recordings from only three muscles (e.g., soleus, semimembranosus, and vastus lateralis) could be a viable option for gait analyses in clinical settings.