Advanced Science Research

Daniel a.

Project title: X-ray Polarization of the Reflection Spectrum with IXPE Data from Cygnus X-1  

Abstract: Black holes are extremely massive objects that are some of the strangest known things in the universe. Black holes have disks of hot gas around them that emit light. There is also a cloud of hotter material near the black hole called the corona that energizes some of the light from the accretion disk. We don’t know the exact shape of this corona and where exactly it is located in relation to the black hole and the accretion disk. 

I attempted to help reveal some of this by using data from a telescope that can measure the polarization of X-rays from black holes.  Polarization describes the direction in which a light wave oscillates. By fitting the polarization data with new models we can account for the shape of the corona. Prior research found that the corona is horizontally extended in the disk’s plane, but it did not account for the contribution to the polarization made by some of the light that is reflected off the disk after being energized in the corona. My project accounts for this using a new model called kynstokes and finds that the corona is still likely extended horizontally. 

Mila a.

Project title: The Disproportionate Effect of the Covid-19 Lockdown on Girls Math Education and Implications for the Gender Gap in Mathematics Careers  

Abstract: Previous research has shown that there is a significant gender gap in higher mathematics degrees and mathematics-based careers, meaning it is still a male-dominated field. Research has also found that high school math grades indicate whether or not a student will pursue a math career, and that the COVID-19 lockdown negatively affected math education in the United States. The impact of the lockdown on female vs male students has not been seriously studied, so we do not know if it has worsened the gender gap in test scores. If the gender gap did get worse, this might affect the gender gap in math careers when this generation reaches college and post-grad stages. 

I analyzed math test scores of 8th graders separated by gender from before and after the lockdown to see if it impacted the gender gap. I found that the lockdown affected the gender gap in mathematics scores when analyzing both average test scores and high-scoring individuals, who are more likely to pursue a career in mathematics. If this increase in the gender gap persists, there may be a need for future educational interventions, which should be explored in future research.  

Sophie e.

Project title: Inner Speech Classification for EEG Brain-Computer Interface: A Comparison of Feature Extraction Techniques 

Abstract: I often exaggerate that I have gained the ability to read minds. In reality, I conducted research on brain-computer interfaces (BCIs), which allow people with severe disabilities to control and communicate with their external environments. 

My project focuses on improving non-invasive, natural BCIs. Therefore, I used a publicly accessible dataset consisting of electroencephalogram (EEG) signals recorded using electrodes on a cap, with inner speech (IS) trials. Each IS trial consists of people thinking the words “up,” “down,” “left,” or “right” in Spanish, which I attempted to differentiate between using machine learning. Because EEG can be highly affected by noise (e.g. eye blinks), and IS, while most natural, is hard to differentiate between as language processing is extremely complex, I focused on extracting features or patterns that help discriminate between signals. 

I applied a novel combination of feature extraction techniques on this dataset, and reached an accuracy of 34% (9% above chance level), using only 1.1 seconds of recorded EEG data. Therefore, while my results were not the most accurate in the literature, my methods were still legitimate, as they were the most efficient. I also set a baseline information transfer rate (ITR), which has not yet been measured in IS-EEG-BCIs, but provides a standard quantification of the speed and accuracy of a BCI in real time.

ilan e.

Project title: Site-specific Incorporation of Methylarginine Into Proteins Using a Mutant Leucyl-tRNA Synthetase

Abstract: I am developing a novel method of producing synthetic proteins with a structural modification called arginine methylation. Arginine methylation is significant because its dysregulation can lead to the development of cancer. Obtaining proteins with a certain degree of arginine methylation in specific places is currently impossible inside living cells, but I aim to change that. My method will modify the protein synthesis process by utilizing a novel tRNA-synthetase protein I created based on computational predictions, which would allow the cell to add methylarginine into proteins when it normally would be unable to do so.

To create designs for the novel protein, I used an AI-based protein prediction software developed by Google called AlphaFold. I based the structures of the new proteins on existing proteins that interact with methylarginine. 

Using this software, I created 2 possible designs for candidate proteins. My work aims to help scientists develop treatments for cancer and investigate arginine methylation by allowing them to acquire proteins for research more easily.

Dylan f.

Project title: TREM2-Fc’s impact on tau in 4R tau h-iPSC neuronal model of Alzheimer’s Disease

Abstract: In the brain, Alzheimer's Disease disrupts processes vital to neurons, including communication, and metabolism, causing neuronal death. Alzheimer’s Disease is closely associated with tau, a protein regulating neurons from within, which accumulates to form tangles in AD, blocking neurons from performing their functions. Microglia, immune cells in the brain, function due to their receptors: triggering receptors on myeloid cells 2 (TREM2). Soluble TREM2 (sTREM2), a fragment of TREM2, plays an important role in regulating microglial function and can bind to receptors on neurons and influence tau levels, which has been studied in mouse models and humans.

I wondered what the relationship is between different concentrations of sTREM2 and tau in the novel 4R tau human-induced pluripotent stem cell model.  This model is unique in that is it derived from human stem cells that are genetically reprogrammed to act like human neurons. Tau aggregates are then induced in the model in both isoforms, or versions, to create a more accurate model of the human AD brain. 

I worked the 4R tau h-iPSC model and induced increased sTREM2 in my first culture and less sTREM2 in my second (20nM and 2nM). I also included two models with different timepoints of introducing sTREM2: in Model 1, sTREM2 was added 2 weeks post-induction of tau seed. For Model 2, sTREM2 was added simultaneous to the tau seed. Then, I compared the levels of tau in for each experimental group and the controls using immunofluorescence staining and Western Blot. These two methodologies detect aggregated tau.

I found that in general, the addition of sTREM2 decreased aggregated tau levels in Model 1 but not Model 2, especially in the cultures with more sTREM2 (20nM). This shows a time effect in that sTREM2 might reduce already-aggregated tau and have less of an effect on tau seeds as they are forming aggregates. This study is significant because it supports past studies using different models and reveals sTREM2 as a therapy target to alter tau inside AD neurons. This novel model is the only platform where the pathological aggregation of tau in a dish can be created, so it adds a new human-based piece of data to our knowledge of the relationship between sTREM2 and tau aggregation.

sofia g.

Project title: Age-Related Differences in Susceptibility to Visual and Auditory Stimuli in Students from 3rd to 12th Grade

Abstract: Schizophrenia is an acute mental illness that is characterized in part by visual, as well as auditory hallucinations, which can be extremely debilitating to the individual depending on the severity of the diagnosis. Schizophrenia’s symptoms are categorized into two main types. The first are positive symptoms, which include hallucinations, delusions, and disorganized thinking. The second is negative symptoms, which involve diminished emotional expression, social withdrawal, and a lack of motivation or pleasure in everyday activities. This illness affects approximately 24 million individuals worldwide, and 2.8 million adults in the United States. 

While these hallucinations have been vastly studied in patients with schizophrenia, there is a knowledge gap regarding the underlying perceptual processes that are less understood, especially how these processes differ from those in healthy individuals. For example, visual and auditory pathways are crucial for accurate emotion recognition in healthy development. These skills are essential for social interaction and are typically robust by adolescence. Past research has shown that deficits in auditory emotion recognition, and facial emotion perception are both core features of schizophrenia, as well as key components of social cognitive impairment. These deficits in emotion recognition could also be translated to poor ability to recognize tonal or prosodic features of speech that convey emotion. This would include changes in base pitch and pitch variability. 

To understand this issue, I studied how visual and auditory processes develop in healthy children and adolescents, and this became the baseline that helped me understand the differences between patients with schizophrenia and healthy controls. Furthermore, this helped me understand how visual and auditory pathways develop in patients with schizophrenia and differ from healthy development, leading to hallucinations or psychosis later on in development. 

Max s.

Project title: Developing RNA-Targeting Tools to Dissect the Function of lncRNA ANRIL

Abstract: Heart disease is a leading cause of death in the world, so it is important to combat it. While some kinds of heart disease are preventable, genetic heart disease is not. Researchers found that ANRIL is a gene that causes up to 15% of all of this genetic heart disease. ANRIL comes in two forms, long and short with the short form being connected to heart disease. Previously the DNA that codes for it has been destroyed, which prevents heart disease, but in a real person this is not possible. 

In this study I targetted the RNA for ANRIL, which in the past has been possible to suppress in a real human. I did this in human kidney cells, in order to determine if it is possible at all to suppress ANRIL. I made genetic code to help my Antisense Oligonucleotides (a method of finding and destroying RNA) locate the ANRIL, after this I transfected the ASOs into the HEK293T cells with ANRIL being expressed. I found that it is possible to knockdown the short ANRIL RNA, which is relevant because now I can move towards targetting ANRIL in heart cells where it is needed most.

jada s.

Project title: Using a DCNN model with different parameters and an LSTM layer to select attention in a multi-speaker scenario

Abstract: Humans have the incredible ability to focus on one speaker in a crowded setting with multiple speakers, such as a cocktail party. However, people who are hearing-impaired cannot perform this task, even with hearing devices. This is called the cocktail party problem. Recent studies have attempted to solve this problem by using neural network models. These models would process the speech envelopes of the speakers in the environment and the EEG of the listener to then compare them and identify the attended speaker by the speech envelope that corresponds with the EEG. 

I aimed to increase the accuracy of the dilated convolutional neural network (DCNN) model from Accou et al. (2021), and understand how parameters affected the model performance. In my study, I used the DCNN model from Accou et al. (2021) and my proposed model of the DCNN model with a long-short term memory (LSTM) layer. 

I found that there was a positive correlation between the window length and accuracy, meanwhile, there was a negative correlation between the number of mismatches and accuracy. Further research can be done on the models and on other ways to increase the accuracy. The DCNN-LSTM model can be tested with different parameters such as more LSTM layers to see the performance capabilities of this model. Improvements and different changes to the model architecture can increase the accuracy of the model with multiple mismatches. Through further development of the neural network models, there can be a hearing device that solves the cocktail party problem.  

Phoebe C.

Project title: The Impact of Cyclone Pam on Birth Weight in Vanuatu 

Abstract: There are 32 million low-birth-weight (LBW) infants born globally every year, and 96% of those LBW infants occur in lower-middle-income countries (LMICs). One factor that affects LBW outcomes is environmental disasters. My study focuses on Cyclone Pam, a category 5 cyclone that hit Vanuatu, a small LMIC in the South Pacific, in March 2015. We know tropical cyclones increase the odds of LBW. Although the connection between natural disasters and increased risk of LBW is established, there is a lack of research conducted in LMIC. To address this gap, I studied the birth records before, during, and after the environmental exposure to cyclone Pam in Vanuatu. 

First, I calculated the mean BW by trimester of exposure to cyclone Pam to determine the overall trend of BW from 2009 to 2016. Unexpectedly, the BW increased in Vanuatu from 2009 to 2016. I used linear regression to confirm the general trend. I found the same increase in BW when comparing pregnant mothers exposed to cyclone Pam in trimester 1 to mothers exposed to cyclone Pam in preconception. The upward trend in BW exemplifies a general increase in maternal care in Vanuatu. This is especially important as Vanuatu’s location is particularly vulnerable to natural disasters, which are expected to increase in severity and frequency due to climate change.

To try to explain this pattern, I analyzed maternal hemoglobin levels (Hb) before, during, and after cyclone Pam, as moderate levels of Hb had been found to indicate a healthy and increasing BW trend. There were no significant differences in maternal Hb levels during, before, and after cyclone Pam when compared to the same months several years earlier. Given that my research cannot connect increasing BW with Hb levels during cyclone Pam, future research should analyze if maternal Hb is affected by natural disasters and if maternal Hb affects BW outcomes. 

This study provides hope for other LMICs that it’s possible to improve prenatal care and overcome the risk of LBW when faced with severe natural disasters.

Mia G.

Project title: Discovering the causal variants in dyslexia through combining genetic and neuroscience studies

Abstract: Dyslexia, which affects approximately 5-12% of English speakers, isa complex learning disability characterized by challenges in reading and interpreting words. The landscape of genetic and neuroimaging studies on dyslexia is marked by inconsistencies and potential false positives, impeding our understanding of the genetic mechanisms underlying dyslexia.

My research aims to address these challenges through a methodology that involves comparing genetic and neuroscience studies. I used a neuroscience study provided to me by my mentor, which looks at the variants that have an effect on the brain. This choice was based on the premise that variants leading to gene differences, and subsequently differences in brain phenotype, contribute to the behavioral presentation of dyslexia. To ensure the appropriateness of eQTL data, I compared different types of brain data with the dyslexia dataset that I received through 23andme by emailing the authors, and identifying the most shared variants between the eQTL and GWAS datasets.

I ran my analysis to identify causal variants in dyslexia, which considered linkage disequilibrium (LD). LD we used was in European ancestry to pinpoint genetic variants likely to be causal. This analysis, which looks at the heritability between variants on a scale of 0 to 1, holds the potential to identify dyslexia-linked genes that were not statistically significant in other studies.

The outcomes of my study unveiled chromosome 9 as a previously undiscovered and not statistically significant region associated with dyslexia. Importantly, the identified regions I found (on chromosome 9) also impacts the cerebellar hemisphere, which is thought to be one of the underlying causes of reading difficulties and learning disabilities like dyslexia. It's worth noting that despite some dyslexia genetic studies pointing to variants on chromosome 9, they have not achieved statistical significance until now.

Mysha  J.

Project title: The association of the traits of urban aquatic birds and their population genetic patterns

Abstract: As urban development expands, the conditions presented by cities play more of a role in the genetic structures of various species of animals and plants. For some species, cities can be beneficial; for example, pigeons are widespread in cities, and the habitat offered by cities fits their needs. However, for other species, urbanization can reduce their genetic diversity. This was made apparent in my mentor’s study, which showed that for non-aquatic birds urbanization benefited their population genetic structures by increasing their genetic diversity, but the opposite was true for aquatic birds. To understand this disparity, I tested the hypothesis that traits, more commonly associated with aquatic birds, impacted the relationship between urbanization and genetic diversity. 

I used a publicly available dataset on bird traits and tested the effect of 13 different traits. Surprisingly, I found that none of them had an association with the population-genetic patterns. While further testing may be needed to completely rule out this hypothesis, it may also be true that other factors such as water pollution in urban spaces may be a more viable explanation for this disparity.

Maya K.

Project title: Validating a liquid culture as a control group for 2D and 3D leukemia cell models

Abstract: Leukemia is characterized by the expansion of abnormal blood cells in the bone marrow and blood. Leukemia cells are surrounded and altered by their microenvironment, which protects and provides a sanctuary to them. This helps them evade common treatment options such as chemotherapy, which can ultimately lead to relapse for many patients. The goal of my study was to validate the use of a liquid culture as a reference point for a new 3D model that my mentor and his lab are developing and using to evaluate different leukemia cell lines in the context of their microenvironment. 

After analyzing the amount of cells present in the liquid culture with a Cell Countess machine in the lab, I found that the the cells in the liquid culture increased on each day, showing that the cells were not proliferating at a reduced rate and behaved as expected. In addition, at the end of the three days, there were more leukemia cells present in the liquid culture compared to the 2D and 3D cultures, meaning that cells in the liquid culture proliferated at a higher rate than cells in the other models. 

My project validated the use of a liquid control as a control group for 2D and 3D leukemia cell models, and showed certain advantages to the use of the 3D model. With a valid control and confirmation that certain newer models are better able to replicate the bone marrow microenvironment, more studies should investigate 3D models and their characteristics for evaluating leukemia, to eventually be able to study the interactions between leukemia cells and their bone marrow microenvironment in 3D, potentially with the addition of treatment. 

Mira L.

Project title: Identifying mechanisms of lysosomal cross-correction in MPSIIIC by gene-modified hematopoietic stem and progenitor cells

Abstract: Mucopolysaccharidosis type IIIC (MPSIIIC) is a severe neurological disorder that affects lysosomal enzyme HGSNAT and impacts its ability to break down a complex sugar molecule called heparan sulfate. To treat this disease, my mentor proposed using a gene-modified stem cell therapy. They found that the treatment effectively reduces the accumulation of heparan sulfate within cells, suggesting healthy enzymes are being transferred from stem cells to diseased cells, which allows for the breakdown of heparan sulfate. However, they did not know how this process is occurring. My goal was to investigate the process by which healthy copies of the HGSNAT enzyme are transferred to disease cells. Specifically, I wanted to investigate the role of tunneling nanotubes (TNTs), which are tiny tube-like structures that form between cells.

At my mentor's lab, I set up cell cultures with diseased cells and healthy stem-cell-derived cells to observe their interactions and mechanisms of enzyme transfer. Once I set up cocultures, I used time-lapse imaging to observe the subsequent cellular communication. 

I found that TNTs enabled HGSNAT transfer from healthy to enzyme-deficient diseased cells. Additionally, I identified the fusion of healthy and diseased lysosomes within diseased cells as a means of alleviating sugar buildup. This is the first report of TNT-mediated transfer of healthy enzymes to diseased cells in the context of a neurodegenerative disease after this kind of a stem cell treatment. These results overall contribute to the understanding of HSPC treatment as well as the possibility for FDA approval of this treatment for human patients, which required some understanding of the mechanism. 

Grey L.

Project title: Abundant preserved foraminiferal organic linings at Piermont Pier (Hudson River Estuary) indicative of benthic ecosystem health, and potentially Trochammina Inflata 

Abstract: I investigated Golden Foraminiferal Organic Linings (GFOLs) in the Hudson River Estuary at Piermont Pier in Piermont, NY. The goal was to identify them at the genus level and understand the environmental conditions related to their abundance. Foraminifera are single-celled microorganisms that have an outer test (shell) that rarely becomes separated from the inner organic lining, which is essentially a smaller, internal, shell. There is a lack of a GFOL database as foraminifera are primarily speciated by their test morphology, and separated GFOLs lack their tests. Specific environmental conditions that lead to the preservation of GFOLs are also unknown.

I collected data in three primary ways: (a) by assessing the amount of organic matter in a sample by weighing sediment samples before and after burning them, to measure how much organic matter had burned off; (b) by using X-ray fluorescence to analyze the elements present within the sediment; and (c) by counting and identifying the abundances of the six species of foraminifera that my mentor's lab was looking for, along with counting and identifying the GFOLs, all of which I did by examining the samples under a microscope. 

I found that (1) GFOLs and another foraminifera species, T. inflata, are correlated with the amount of organic matter in a sample; (2) T. inflata and GFOLs are correlated with each other; (3) Two T. inflata specimens from all picked foraminifera contained visible GFOLs in the center of their spiral. The GFOLs were correlated with metals manganese and rubidium, both indicators of human pollution. These three converging pieces of evidence suggest that the GFOLs found are the GFOLs of T. inflata that had lost their tests due to high amounts of organic matter, manganese, and rubidium. 

Michael S.

Project title: Fast and efficient ion transport in a trapped-ion quantum computer through low-pass filter design

Abstract: The field of quantum computing is the application of quantum physics to create a computer that is faster and more efficient than any other computer. A trapped-ion quantum computer is a type of quantum computer that uses trapped ions as the base piece of information known as a qubit. Bits and qubits are the basic units of structure underlying all computers. Performing operations with multiple qubits can be successful only if the transport of the ion is done fast and efficiently. The problem is that electronic signals that are meant to move the ion within the trap often contain the kind of noise that can eject the ion out of the trap. This noise needs to be filtered out. 

To achieve this, I designed 6 different filters and then simulated those 6 different filters based on two different types of signals: triangle, and ion transport. The triangle signals were used to assess the general structure of each of the filters, and the ion transport signals were used to determine which of the filters would be the best fit for the ion trap. I used a program called LTspice to simulate the filters, and I imported the data for both simulations into Mathematica. I analyzed the data based on how much motion the ion gained throughout the transport process. The simulations were also performed on a range of transport durations to assess the effects at a fast transport time.  

I found that the Butterworth 5-pole is the best filter for triangle signals and the Bessel 3-pole is the best for fast ion transport. The implementation of the most successful low-pass filter from the second set of tests in the ion trap system will result in a general improvement in the transport operation in a trapped ion quantum computer. This improvement in transport will subsequently improve the overall performance of the quantum computer when performing operations with multiple pieces of information (i.e., multiple ions).

Graham S. 

Project title: Performance prediction for future VTOL unmanned aerial vehicles using preliminary sizing models

Abstract: The use of battery-powered Unmanned Aerial Vehicles (UAVs, a.k.a “drones”) is increasing and will continue to increase as battery technology improves. The problem is that it is unclear how these technological advancements will affect future UAV performance. Therefore, the goal of my project was to predict how flight range and other performance characteristics of future UAVs will be affected by better battery technology. 

I did this by changing a commonly used plane design tool for fuel-burning planes to work with electric UAVs. To do this, I manually collected data online on 40 UAV designs as there was no dataset for the designs and specifications that I needed, which was used for training the design tool. Then, I determined the accuracy of my UAV design tool by comparing it to items in my database and comparing it to a similar model for regular planes. 

My tool was slightly less accurate than the comparable model, with an 18 percentage point higher error, which was considered a good result considering the lower amount of data relating to electric drones. I then varied the different battery technologies and inputted the assumed values into my tool to predict the future performance of drones. In summary, I made a uniquely computationally simple and general UAV design tool.

Gavin Y. 

Project title: De novo drug design as GPT language modeling: Large chemistry models with supervised and reinforcement learning

Abstract: Drug discovery is one of the most time-consuming and costly aspects of developing a drug. It is estimated to take about 10–15 years, with a cost of $1.4 billion per drug discovered and approved. In simpler terms, it is impossible to enumerate all possible synthesizable molecules to test for potential drug effectiveness. Machine learning (ML) has emerged as one of the most promising tools in drug discovery and can speed up this process. Because molecules can be represented in “languages” that an algorithm can interpret, it is possible to retool language processing ML models such as GPT models for drug design. Traditional non-large language models for drug design from previous studies often generate nonsensical (invalid) representations that do not represent actual molecules (like ChatGPT producing gibberish). Thus, my goal was to use GPT to design drug candidates that are both highly effective in targeting a specified drug target and chemically valid.

Before I could train a GPT model for generating effective molecules, I needed a way to evaluate the drug effectiveness (a.k.a efficacy) of any molecule quickly, as it would be impractical to synthesize every machine-designed molecule throughout the entire training process. Thus, I designed my own efficacy evaluation model. Then, I trained my GPT drug design model to design similar drug-like molecules as candidates designed by humans from the dataset. Finally, I used my trained efficacy evaluation model to optimize my drug design GPT model for designing higher efficacy molecules using reinforcement learning. I used the amyloid-precursor protein (APP), a promising drug target for Alzheimer's disease, as a case study for using GPT for drug design. However, the same methodology can be applied to transfer my GPT drug design model for targeting a different protein using a different dataset.

My drug efficacy evaluation model is 2.3 times better in accuracy and is more data efficient compared to models from previous studies. Almost all (99.2%) of the designed molecules by my drug design model are highly effective, and almost all of them are more effective than the average effectiveness of the molecules from the training dataset. In addition, all of the designed molecules are still chemically valid, and novel, as the designed molecules do not exist in the dataset. Future studies can take inspiration from OpenAI’s methods and have human chemists provide feedback directly to the drug design model. The GPT-designed drug candidate molecules exhibit similar properties as the ones from the dataset, and thus future studies can leverage this phenomenon to make patented drugs accessible by generating similar ones. All of these implications from results and future research extensions have the potential to transform drug discovery.

2024 Symposium recording and poster session

2024 Symposium PROGRAM

"At the beginning of high school, I knew I wanted to do STEM in the future, but I had no idea what it would be really like, and I didn't know what field I wanted to pursue. With ASR, I got the chance to actually see what STEM would be like in the real world, and I really loved it! The work I did in ASR was the most meaningful work I did in high school since I got to focus on what I was really interested in, and it gave me a lot of research skills that I utilize now in college for both research and classes."

Alex Lyons '21

"ASR is just as much about training you how to think and study as about how to be a scientist. Besides introducing you to the world of science and how it is done, ASR also continuously pushes you to improve yourself, to study and manage your time more efficiently. ASR is about learning how to tackle seemingly impossible challenges; it teaches you how to break them down into tangible and achievable pieces and work on them one at a time. Ultimately, ASR will leave you with a problem-solving approach that will stick with you even after you leave the program."

Yuqiao Zou '22

"ASR was a powerful influence on both Annika’s education and her life’s goals; the class provided a combination of timely, relevant knowledge on a topic that’s critical to society and can impact our future. The subject matter of the class had a game-changing effect, prompting her decision to make science the focus of her career. As parents, we highly recommend this class as we watched it build our daughter's character and confidence while broadening her understanding and perspective on current events."

Jon and Becky Spaet, parents of annika '23

"Although the School has many strong disciplines, ASR was by far the most beneficial experience our son [Alex] had at CGPS. Due in large part to the talented and dedicated faculty, the program literally changed the course of his young life. He learned how to best approach complex and long-term projects, developed self reliance and resilience, and discovered and pursued his intellectual passion. We couldn’t recommend this program more!"

Josefina and Gregory Lyons, parents of alex '21

"We are so grateful for Mr. Yashin and the ASR program.  It has been an incredible all around experience for our son that has also served to benefit him in all of his classes. We have seen a noticeable improvement in his communications and writing skills which has made him a confident presenter in and out of ASR.  Lastly, his interaction with both Mr. Yashin and his project mentor has been invaluable as he has been exposed to and learned things that he likely would not have been exposed to until college."

Jill and Jeff Libshutz, parents of lucas '23

"ASR is hard work, but it pays off in spades. We have seen tremendous growth in Arjun. In his research journey, he progressed from knowing very little about healthcare and struggling to understand journal articles to authoring a published paper on the subject. The structure of the program facilitates growth as it pairs individual responsibility with mentoring and support. He has learned to organize his time and pace his work to meet commitments. Research isn't always easy, and Arjun has faced his share of setbacks, but learning to cope with them is an important skill he has acquired, and this experience has made him resilient. In addition to the domain knowledge he has accumulated, the process of writing a lengthy review article and scientific paper and going through numerous revisions and presentations has led to a strong improvement in his writing and presentation skills. The close-knit community where the students help and support each other is a highlight of his experience and has reinforced in him that success is not an individual effort but a team sport. If you are curious about science and desire to make an impact, ASR provides an ideal platform to realize your ambitions.

Arjun had the opportunity to join other selective schools but was drawn by the unique opportunity at CGPS to conduct college-level scientific research on a topic of his choosing in the ASR program. We can confidently say that the mentorship and support that ASR provided enabled him to advance his research much further than he might have been able to at other schools. Judging by the remarkable growth in his skills, knowledge, and confidence in a relatively short time by being part of ASR, we feel that we made the right decision."

Roopali and Ushane Sharma, parents of arjun '23