GeniCare AI

Currently, more than 480 million individuals globally, representing over 6.2% of the world's population, struggle with rare genetic disorders. In the United States alone, over 7,000 distinct rare diseases collectively inflict over 30 million individuals. The intricate and diverse nature of these genetic disorders presents substantial challenges in achieving accurate diagnoses, leading to an extended and often frustrating diagnostic journey known as a 'diagnostic odyssey.'

Primary care physicians are confronted with the formidable task of navigating a complex web of clinical conditions, many of which are exceptionally rare and manifest with atypical and perplexing symptoms. The diagnostic process is further complicated by the need to distinguish among numerous potential rare diseases, a task that frequently demands elusive expertise. Therefore, the delay in getting an eventual genetic diagnosis is often caused by the lack of expertise and knowledge by primary care providers, especially pediatricians who are unable to recognize early signs of genetic syndromes. 

Our team is driven by these intertwined issues and is committed to deploying a multimodal AI aid for assisting primary care providers in the timely identification of patients suspected to have rare genetic diseases, facilitating the referrable to disease specialists for further neurological, behavioral and genetic evaluations.

Our innovative knowledge-based and data-driven AI aid employs a Multimodal approach, leveraging patient facial images, video recordings and comprehensive and longitudinal clinical phenotype data, including demographic information and social determinants of health data, to generate predictive insights for rare genetic diseases. To accomplish this, we synergize advanced Multimodal Machine Learning and graph-based algorithms with cutting-edge Large Language Models (LLM).

Specially,  

• Our Multimodal Machine Learning (MML) algorithms will skillfully extract clinical facial phenotypes from patients' facial images, discerning subtle details. If available, video recordings can also be used to assess behavioral phenotypes and neuromuscular phenotypes that are not present in static photos.
• Subsequently, we will use GPT models and rigorous graph theory algorithms to analyze this valuable information together with patients' clinical notes and demographic specifics, such as gender, age, and ethnicity. 

This comprehensive analysis empowers our solution to identify potential rare diseases accurately and to recommend the next course of action regarding which specialty care facilities should the patient be referred to, which genetic tests or genomic tests shoud be ordered, and so on. 

Our target population is pediatric primary care clinicians who serve pediatric patients, especially those underserved and underrepresented minority groups. Certain minority groups encounter challenges that stem from systemic biases ingrained within data collection and analysis processes. These biases can inadvertently lead to misrepresentations, inaccuracies, and disparities in the rare genetic disorder predictions of these groups.

In addition to multimodal data, our machine learning model incorporates patients' demographic characteristics during training, effectively mitigating bias and enhancing prediction accuracy. As a result, it ensures equitable diagnoses for all patients, particularly in the context of addressing rare genetic diseases among minority populations.

Our team prioritizes a customer-centric approach in model development. We continuously engage in interviews with primary care pediatricians to gain a deep understanding of their genuine requirements. This encompasses their specific interests in diseases, the patient groups they frequently encounter, and the user interface design that can facilitate physician ease of use. 

To illustrate, one clinician aimed to differentiate between two distinct syndromes. As a result, we rigorously educated our model by incorporating real patients’ data and medical literature, equipping it to competently handle this specific diagnostic challenge. Furthermore, another medical expert exhibited curiosity about a specific medical condition, prompting us to gather and integrate the most recent patient cases into our model to meet their specific requirements.

Through these interviews and model refinements, we ensure that our startup's product is firmly rooted in reality, developing AI solutions that can genuinely create a tangible impact.

In a world where the quest for precision and equity in diagnosis is paramount, we've harnessed the power of Multimodal AI, merging Large Language Models with cutting-edge graph-based algorithms to improve the diagnostic process. Traditionally, AI solutions in this domain have fixated on facial images as the primary data source for model improvement. While facial images can be informative, they often fall short in delivering a comprehensive diagnostic picture, especially for conditions like Noonan syndrome (NS), Prader-Willi syndrome (PWS), Silver-Russell syndrome (SRS), and Aarskog-Scott syndrome (ASS), which exhibit shared characteristics such as short stature. Furthermore, crucial indicators like sleep disturbances, impaired balance, and intellectual disabilities remain elusive through facial images alone. However, some of these features can be represented by video recordings, and often by clinical narratives written by primary care physicians.

Acknowledging the gap, we adopt a different approach with diverse data inputs. We do not confine ourselves to facial images alone; rather, we encompass a holistic spectrum of data sources, including demographic information and clinical notes, and potentially many more.

One of the keystones of our innovation lies in the integration of demographic data into the predictive model. Factors like gender, age, and ethnicity are pivotal elements of a patient's medical profile. By incorporating these elements into the textual inputs, our model gains the ability to discern unique patterns associated with each rare disorder. In doing so, we address the potential biases that might lurk within the data, especially concerning underrepresented minority groups. Our commitment to fairness ensures that every patient, regardless of background, receives an equitable diagnostic assessment.

The innovation doesn't stop there. We've embraced the power of graph-based algorithms to enhance our predictive accuracy. These algorithms offer an interpretive lens through which we can elucidate results in a clear and understandable manner, fostering trust and confidence in our diagnostic process.

In addition, our solution leverages the capabilities of Large Language Models, which bring a linguistic finesse to our interactions with pediatricians. This user-friendly approach ensures effective communication, making the diagnostic process more collaborative and accessible. The "few-shot" learning capabilities of our model also enable accurate predictions even with limited training data, ensuring adaptability and precision in various clinical settings.

We're not bound by a single modality of data. Our solution's versatility extends to seamlessly incorporating additional data modalities such as audio and video. This adaptability is key to our commitment to continuous improvement and ongoing model refinement.

By transcending the confines of facial images, integrating demographic data, and embracing state-of-the-art algorithms and Large Language Models, we're poised to redefine the future of rare genetic disorder diagnosis. Our startup's unwavering commitment to delivering impactful and equitable AI solutions is grounded in reality, bringing tangible benefits to both patients and healthcare providers.

Our Multimodal AI solutions have three primary objectives, all of which are in alignment with UN Sustainable Development Goal 3 - Good Health and Well-Being:

(1): Achieve swift, precise, and effective diagnosis for patients, with a particular focus on pediatric cases involving potential rare diseases. Leveraging the capabilities of advanced AI models and multiple data sources, we strive to excel in delivering near-perfect diagnostic outcomes.

(2): Significantly lower the financial burdens associated with medical diagnosis, with a particular emphasis on underserved populations grappling with poverty and limited access to essential healthcare services. Through systematic and algorithmic disease diagnosis, we aim to reduce unnecessary costs. By operating our software with utmost cost efficiency, we aspire to make our services highly affordable, with a strong commitment to providing free assistance to those living in impoverished conditions.

(3): Promote equitable diagnostic practices that transcend demographic disparities. Our algorithm is purposefully crafted to confront and mitigate potential biases within medical data, especially those affecting underrepresented minority groups. Our unwavering dedication to fairness ensures that every patient, regardless of their background, receives an impartial and just diagnostic evaluation.

Our current datasets, as detailed below, encompass a variety of sources for our research and AI development:

• Publicly available databases: These databases compile facial images of patients along with their associated clinical characteristics, sourced from medical publications with due patient consent. We have access to approximately 7,500 images, with ongoing efforts to reach 8,000.
• OMIM/OrphaNet Database: publicly accessible database that comprehensively documents clinical features, genetic information, and other pertinent data related to nearly all rare genetic diseases.
• Newly Published Case Studies: This dataset encompasses case studies published from 2022 onwards. While these cases are not presently part of the publicly available database, they are obtainable for use with the explicit consent of the patients involved.
• Private Patient Data from Clinicians: This dataset represents a unique source of exceptionally high-quality data contributed by in-house experts with appropriate IRB protocols.

It's important to note that only dataset (4) comprises private data sources, while all others are publicly available for research and development.

Our AI solutions can be divided into three distinct components:

(1): Multimodal AI: This module is designed to receive patients' facial images as inputs and generate assessments of abnormal facial characteristics, presenting them in the form of clinical phenotypes.

(2): Embedding Phenotypes of Patients into Weighted Human Phenotype Ontology (HPO) Graphs: In this component, we embed the identified phenotypes derived from both facial images and real clinical notes in the form of subgraph of (weighted) Human Phenotype Ontology (HPO). Notice that the weighted HPO graph serves to underscore the significance of each phenotype. Importantly, the weights assigned to each phenotype are adjusted to accommodate variations across different demographic groups, ensuring a more precise representation of the underlying data.

(3): Large Language Models-Based Predictions: Leveraging the embedded graphs, our Large Language Models undertake a comparison of similarity with referenced “true” graph structures, extracted from the OMIM database. Subsequently, the system makes predictions, provides a rationale for its decisions through a weighted graph, and ultimately suggests a follow-up treatment plan.

Ethical and Responsibility Considerations:

1. Privacy and Security Safeguards: (1) The majority of our data sources are publicly available and obtained with patients' informed consent, ensuring safe and ethical use. (2) Our in-house private data is analyzed within the secure Cloud computing system, preventing data leakage. (3) Our AI solutions are designed to focus on learning relationships between phenotype patterns and disease predictions, rather than personal patient information. The deployment of our models ensures the protection of personal data, with outputs limited to phenotypes during the multimodal AI stage, and diseases predictions, graph-based rationales, and treatment plans in later stages.
2. Compliance with Regulatory Standards:

• We are fully committed to complying with governmental regulations that uphold transparency, ethics, and patient well-being.

3. Algorithm Transparency and Accountability:

• While we maintain a competitive edge in the market by not disclosing proprietary algorithms, our solutions can be made fully transparent and interpretable to enable healthcare professionals and policymakers to understand decision-making processes.

4. Ethical Guidelines and Standards:

• Our databases and models are subject to continuous management and updates to ensure the comprehensive protection of patients' rights.

5. Affordability and Accessibility Initiatives:

• We are open to collaborating with policymakers to make our AI models cost-effective and accessible, particularly for underserved populations.

In our commitment to ethical and responsible AI development, we prioritize patient privacy, security, transparency, and compliance with all relevant regulations. These principles underpin our mission to deliver AI solutions that benefit individuals, healthcare professionals, and society at large.

In the upcoming year, our aspiration is to introduce our products to primary care facilities located in Philadelphia and New York City, ultimately enriching the experience of pediatricians with our AI-assisted solutions. In parallel, we look forward to continuing refinement of our prototype according to their valuable feedback and constructive critiques related to reliability, user experience, and suggestions for enhancement.

In the next five years, our impact goals are to 

(1) Expand to Major Hospitals in the Northeastern United States: Our vision is to extend our reach to some of the most prominent medical institutions in the region. By doing so, we aim to provide cutting-edge AI-assisted solutions to a broader range of healthcare providers, enhancing diagnostic accuracy and patient care.

(2) Empower Underserved Populations Within the United States:

• Numerically, our goal is to directly impact the lives of at least 2-3 million underserved individuals, especially children, in the United States. This could include low-income communities, marginalized groups, or those residing in healthcare deserts.
• We intend to collaborate with community health centers, local clinics, and non-profit organizations to ensure that AI-driven healthcare solutions are accessible to those who need them the most.

(3) Extend Support to Underserved Global Communities:

• Our overarching objective is to make a tangible difference in the lives of at least 100 million underserved individuals worldwide. This may encompass impoverished regions, remote communities, and vulnerable populations with limited access to healthcare.
• Through strategic partnerships with international healthcare organizations and NGOs, we aspire to deploy our AI solutions to address healthcare disparities on a global scale.

Four people total. Two full-time staff, two part-time.

We have spent over one year on this solution.

Our team comprises a dynamic group of interdisciplinary researchers, each with diverse research interests and expertise that spans fields such as bioinformatics, probability theory, statistics, topology, and more. We are steadfast in our quest to welcome individuals from varied backgrounds who share our core values and vision. Our collective aim is to develop proficient, responsible, equitable, and accessible AI products tailored for primary care healthcare professionals.

We firmly believe that the path to creating a successful and impactful AI product necessitates ongoing collaboration with healthcare professionals. Furthermore, we recognize the vital role of business expertise in driving the success of our startup. As part of our commitment to achieving these objectives, we are actively engaged in initiatives such as the Mid-Atlantic I-Corps program and Nucleate Philadelphia.

In these programs, we seek to establish meaningful connections with seasoned healthcare professionals capable of offering constructive critiques of our product, as well as experienced business professionals who can provide essential guidance on the entrepreneurial aspects of our endeavor. Should they choose to join our cause, we welcome the possibility of recruiting them into our leadership team.

Our journey has already led us to connect with clinicians whose invaluable insights have enriched our model. This experience underscores the power of diversity in thought and expertise. It is this belief in the strength of diverse perspectives that drives our participation in challenges like this one, where we aspire to engage with individuals from varied backgrounds and extend invitations for collaboration.

Our team is composed of four key members. Dr. Jeremy Wu assumes the role of team lead and is primarily responsible for spearheading the main algorithm design. Dr. Jingye Yang and Dr. Jeremy Wu are entrusted with the main tasks of main algorithm implementation, interviewing customers, and product development. Dr. Kai Wang and Dr. Chunhua Weng provide project oversight and secure project funding.

We have successfully developed a prototype capable of diagnosing up to 450 rare diseases with remarkable accuracy. Currently, we are actively engaged in the Mid-Atlantic I-Corps program, where our primary focus is conducting in-depth interviews with primary care pediatricians—our main customers and vital stakeholders. This process serves a dual purpose: to collect invaluable feedback and suggestions while simultaneously fine-tuning and perfecting our model. This phase is anticipated to extend over the next a few months, with a commitment to ongoing refinement as required.

Building on the insights gained through these engagements, we are in the process of gathering high-quality patient data from in-house datasets to further enhance the model's capacity to distinguish specific diseases. Concurrently, we are dedicated to enhancing the model's interpretability by leveraging advanced graph-based algorithms. Our overarching objective is to align our model's decision-making rationale more closely with that of human clinicians, thereby embodying the essence of artificial intelligence inspired by human intelligence.

Upon reaching a point where our prototype has been meticulously refined and is primed for beta user testing, we will shift our focus to engaging with an array of other key stakeholders. These include venture capital funds, investors participating in programs like NSF I-Corps, Nucleate, and government regulators, among others. Simultaneously, we will sustain our partnerships with primary care facilities to ensure continued improvements.

Upon establishing strong relationships with key customers, successfully navigating regulatory considerations, and achieving the desired level of product maturity, we will proceed to the formal launch of our startup. This strategic approach ensures that we enter the market well-prepared, having garnered the support and confidence of our stakeholders and poised to make a lasting impact in the field of primary care diagnostics.

At present, the team members are all employees of universities and hospitals. Our primary sources of funding are institutional funds and federal grants, and they will continue to serve as our main financial support until we secure external funding as a separate entity. In addition, we are proactively exploring funding opportunities offered by regional and national entrepreneurship programs, venture capital funds, and similar avenues during the later stages of model development.

Our business model is designed as a Business-to-Business (B2B) model. Following the formal launch of our startup, we project that our revenue stream will be primarily derived from recurring subscription fees paid by primary care facilities. We may also partner with payers, including insurance companies, to cover the recurring subscription fees.

Current operating cost: 

Computing costs + travel costs + fees for interviewing customer: 30K

Projected operating costs for next year: 100K (Details will be listed on the next question)

100K. 

Consulting costs (we will interview multiple consultants for the projects to refine the project goal and needs): 30K

Computing costs, travel cost, miscellaneous cost: 30K

Human capital: we will recruit part-time contractors for development 30K

Legal cost, shipping cost and business supplies: 10K

We are keenly aware of the significant market potential inherent in our current prototype, particularly in its ability to bring tangible enhancements to the existing landscape of primary care. Our team primarily operates within the academic sphere and has limited exposure to industrial settings. Recognizing the imperative to bridge the gap between our academic research and practical, real-world applications, we consider the Cure Residency program as an invaluable opportunity to achieve this harmonious integration. This program not only aligns seamlessly with our overarching vision but also serves as an enabler, empowering us to translate our groundbreaking concepts into practical solutions that can be readily accessed and employed by healthcare providers, ultimately benefitting patients.

Through our participation in the Cure Residency, we aim to achieve the following objectives:

1. Forge Connections with Like-Minded Individuals: This includes fellow entrepreneurs and interdisciplinary experts who share our vision and are eager to collaborate in launching our product, thus collectively contributing to the betterment of our community.
2. Implement a Well-Defined Strategy: We intend to test and launch our prototypes and establish our company with guidance from public health experts and the Cure Executive Advisory Board.
3. Secure Adequate Financial Support: We aspire to access seed funding that will underpin the future development of our prototype.
4. Contribute to the New York Entrepreneurial Community: Our commitment extends to supporting the local entrepreneurial ecosystem through initiatives such as organizing events, sharing insights and experiences, and offering technical assistance.

The most exciting facet of the Cure Residency program lies in the networking opportunities it offers. Our team places great value on fostering meaningful relationships. The mentorship from the Cure Executive Advisory Board represents an invaluable learning resource, enriching our team's understanding of business execution. The chance to connect and learn from kindred spirits in the entrepreneurial realm holds the potential to enhance and refine our strategies. Additionally, the input and feedback provided by healthcare professionals will directly inform the evolution of our prototypes. We firmly believe that our technological proficiency and diverse expertise can also make a valuable contribution to the broader Cure Community.